Загрузка видео...

Не удалось загрузить видео

На главную

Performance Comparison: Uniswap v4 Indexer with the Graph versus SubQuery. Same project. Same setup. The Graph: • Blocks indexed: 135,447 • Duration: 147 mins • Blocks per minute: 921 SubQuery: • Blocks indexed: 295,744 • Duration: 115 mins • Blocks per minute: 2,571 → 2.79x faster Result? SubQuery indexed...

SubQueryNetwork's profile picture

SubQuery

105,007 subscribers

117,588 просмотров • 8 месяцев назад •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

🎉 All tests are passing for Next.js production builds. I asked v0 to visualize this quite literally monumental achievement by the team. 8294 tests passing. Turns out that doing realtime physical simulation over 8294 bodies is quite performance intensive. v0 implemented the same animation in 3 different physics engines (matter-js, planck.js, rapier) effortlessly. In three shots it did what would take a human hours or days. We settled on Rapier which was the fastest. There are many other cool details it nailed. I asked for the falling bodies to increase exponentially. It starts out with a few blocks, then more fall, then it's a torrent of blocks 😁 I'm very happy about our new Turbopack production builds foundation. At Ship we talked about its multi-core aspect, something Rust could give that JS couldn't for a compiler. `next build --turbopack` can actually use the full power of your computer, or your CI system. This has given Vercel single minute compiles on a gigantic codebase. Check it out here:
0:56

🎉 All tests are passing for Next.js production builds. I asked v0 to visualize this quite literally monumental achievement by the team. 8294 tests passing. Turns out that doing realtime physical simulation over 8294 bodies is quite performance intensive. v0 implemented the same animation in 3 different physics engines (matter-js, planck.js, rapier) effortlessly. In three shots it did what would take a human hours or days. We settled on Rapier which was the fastest. There are many other cool details it nailed. I asked for the falling bodies to increase exponentially. It starts out with a few blocks, then more fall, then it's a torrent of blocks 😁 I'm very happy about our new Turbopack production builds foundation. At Ship we talked about its multi-core aspect, something Rust could give that JS couldn't for a compiler. `next build --turbopack` can actually use the full power of your computer, or your CI system. This has given Vercel single minute compiles on a gigantic codebase. Check it out here:

Guillermo Rauch

50,259 просмотров • 1 год назад

In Q2 2024, queries on The Graph Network reached all-time highs following the completion of the Sunrise initiative ☀️ The new record? 3 billion queries in a single quarter 🤯 Additionally, The Graph’s core devs shipped several exciting new features, significantly enhancing the developer experience on the network 🔥 Here are the top 7 key takeaways from the Q2 2024 Participant Update ⬇️ 1️⃣ The Graph Network experienced significant growth, serving nearly 3 billion queries in the quarter—an almost 100% quarter-over-quarter increase. 2️⃣ With the Sunrise of Decentralized Data concluded, over 7,000 subgraphs are now active on The Graph Network. This includes projects like Aave, ens.eth, Art Blocks, Balancer, Snapshot, Decentraland, Lido, Messari, PancakeSwap and Sushi.com. 3️⃣ Scaling via Arbitrum on L2 is officially complete. Queries and fees are now being processed exclusively on L2. 4️⃣ StreamingFast ⏫ launched Codegen, a no-code tool, which prompts developers with questions to auto-generate fully working Substreams. It provides a complete developer environment that remotely compiles the project. 5️⃣ GraphOps | graphops.eth launched GraphSeer, an ecosystem dashboard showcasing Indexer and subgraph deployments data. It’s designed to empower Subgraph Users, Indexers and Delegators to optimize their performance and maximize their impact on the network. 6️⃣ With the help of InfraDAO, new chains continue to integrate with The Graph and complete the official Chain Integration Processes (CIP). Recently integrated chains include: BNB Chain, Base, Linea.eth and Scroll with many more in process. 7️⃣ Bridge Builders DAO continued to provide expert support to web3 builders including supporting in person hackathons at ETH Kyiv and Hack’n Heights. Additionally, BuildersDAO just launched a new website, to better showcase all the ways they can support subgraph developers and The Graph ecosystem. The Graph's Q2 2024 Participant Update shows that with the conclusion of the Sunrise, web3 clearly values a decentralized data layer. Core devs continue to ship meaningful upgrades designed to improve the developer experience and lay a strong foundation for an action packed second half of 2024. Don’t miss any of the details from The Graph’s Q2, 2024. Watch the full update featuring Eva Beylin, 0xMaxTang, , Etienne Brunet, and caro f here:
32:55

In Q2 2024, queries on The Graph Network reached all-time highs following the completion of the Sunrise initiative ☀️ The new record? 3 billion queries in a single quarter 🤯 Additionally, The Graph’s core devs shipped several exciting new features, significantly enhancing the developer experience on the network 🔥 Here are the top 7 key takeaways from the Q2 2024 Participant Update ⬇️ 1️⃣ The Graph Network experienced significant growth, serving nearly 3 billion queries in the quarter—an almost 100% quarter-over-quarter increase. 2️⃣ With the Sunrise of Decentralized Data concluded, over 7,000 subgraphs are now active on The Graph Network. This includes projects like Aave, ens.eth, Art Blocks, Balancer, Snapshot, Decentraland, Lido, Messari, PancakeSwap and Sushi.com. 3️⃣ Scaling via Arbitrum on L2 is officially complete. Queries and fees are now being processed exclusively on L2. 4️⃣ StreamingFast ⏫ launched Codegen, a no-code tool, which prompts developers with questions to auto-generate fully working Substreams. It provides a complete developer environment that remotely compiles the project. 5️⃣ GraphOps | graphops.eth launched GraphSeer, an ecosystem dashboard showcasing Indexer and subgraph deployments data. It’s designed to empower Subgraph Users, Indexers and Delegators to optimize their performance and maximize their impact on the network. 6️⃣ With the help of InfraDAO, new chains continue to integrate with The Graph and complete the official Chain Integration Processes (CIP). Recently integrated chains include: BNB Chain, Base, Linea.eth and Scroll with many more in process. 7️⃣ Bridge Builders DAO continued to provide expert support to web3 builders including supporting in person hackathons at ETH Kyiv and Hack’n Heights. Additionally, BuildersDAO just launched a new website, to better showcase all the ways they can support subgraph developers and The Graph ecosystem. The Graph's Q2 2024 Participant Update shows that with the conclusion of the Sunrise, web3 clearly values a decentralized data layer. Core devs continue to ship meaningful upgrades designed to improve the developer experience and lay a strong foundation for an action packed second half of 2024. Don’t miss any of the details from The Graph’s Q2, 2024. Watch the full update featuring Eva Beylin, 0xMaxTang, , Etienne Brunet, and caro f here:

The Graph

23,429 просмотров • 1 год назад

A study put elite athletes who burned almost no carbs on a treadmill. They recorded the highest fat-burning rate ever measured in a human. The FASTER study. 2016. Published in Metabolism. Twenty of the best ultra-endurance athletes on earth. Ultramarathoners. Ironman triathletes. The kind of people whose careers depend on knowing exactly what fuel works. They were split into two groups. Ten ate the standard high-carb athlete diet. The diet every sports nutritionist still pushes. Eat the carbs. Load the carbs. You cannot perform without the carbs. Ten had been low-carb and keto-adapted for an average of twenty months. Same elite level. Same competitions. No carbs. Both groups ran three hours on a treadmill. Researchers measured exactly what fuel each body was burning, breath by breath. Here is what they found. The keto group burned fat at 2.3 times the rate of the carb group. Peak fat oxidation hit 1.5 grams per minute. The textbooks said the human body maxes out near 0.7 grams per minute. The keto athletes doubled the supposed limit. The highest fat-burning rate ever measured in a human, full stop. Then the part that should have ended the carb-loading dogma forever. The fear was that without carbs they would run out of muscle glycogen mid-race. They did not. Their glycogen use during the run and their refill afterward matched the carb athletes exactly. They were running on their own fat at elite intensity. With glycogen behaving identically. The body was never carb-dependent. It was carb-trained. You can train it differently. Almost a decade later, every endurance handbook still tells athletes to load up on carbs. The data has been sitting there the whole time.
0:57

A study put elite athletes who burned almost no carbs on a treadmill. They recorded the highest fat-burning rate ever measured in a human. The FASTER study. 2016. Published in Metabolism. Twenty of the best ultra-endurance athletes on earth. Ultramarathoners. Ironman triathletes. The kind of people whose careers depend on knowing exactly what fuel works. They were split into two groups. Ten ate the standard high-carb athlete diet. The diet every sports nutritionist still pushes. Eat the carbs. Load the carbs. You cannot perform without the carbs. Ten had been low-carb and keto-adapted for an average of twenty months. Same elite level. Same competitions. No carbs. Both groups ran three hours on a treadmill. Researchers measured exactly what fuel each body was burning, breath by breath. Here is what they found. The keto group burned fat at 2.3 times the rate of the carb group. Peak fat oxidation hit 1.5 grams per minute. The textbooks said the human body maxes out near 0.7 grams per minute. The keto athletes doubled the supposed limit. The highest fat-burning rate ever measured in a human, full stop. Then the part that should have ended the carb-loading dogma forever. The fear was that without carbs they would run out of muscle glycogen mid-race. They did not. Their glycogen use during the run and their refill afterward matched the carb athletes exactly. They were running on their own fat at elite intensity. With glycogen behaving identically. The body was never carb-dependent. It was carb-trained. You can train it differently. Almost a decade later, every endurance handbook still tells athletes to load up on carbs. The data has been sitting there the whole time.

Vinnie Tortorich

116,936 просмотров • 1 месяц назад

A tricky LLM interview question: You're serving a reasoning model on vLLM, and it keeps running out of GPU memory on long traces. So you add KV cache compression and evict 90% of the cached tokens. VRAM usage stays as is and GPU still runs out of memory. Why? (answer below) Evicting 90% of the KV cache can free almost none of the memory it was using. This sounds counterintuitive, but it follows directly from how production servers store the cache today. The KV cache grows with every token a model generates. Each token appends its key and value vectors across every layer, and nothing is freed while generation continues. This is the dominant memory cost for reasoning models. If a 32K-token CoT caches ~32K tokens of KV vectors, a Qwen3-32B with 4-bit weights will run out-of-memory around 24K tokens on a 24GB GPU. One obvious solution is to keep the important tokens and drop the rest, since attention is sparse enough to allow it. But this does not solve the memory problem yet. The reason is paged attention, which is the memory manager behind vLLM and most production servers. Under the hood, it splits GPU memory into fixed physical blocks, each one holds the KV for about 16 tokens. This block returns to the allocator only when every slot inside it is empty. Since the eviction logic selects tokens by importance, and such tokens are scattered across blocks... ...so despite eviction, almost every block is left with at least some survivor tokens. For instance, if the logic evicts 14k of 16k tokens across 1,000 blocks, most likely every block will still have a token. This means the allocator frees almost nothing. Placing the new tokens into those freed slots is not ideal because it breaks the cache's layout. Say token 16,001 arrives, and it's placed in the slot the 40th token used to hold. The cache now reads position 38, then 16,001, then 41, so the cache is no longer in token order. Attention can still compute the right answer from that, but only if every slot now carries a separate note recording which position it actually holds. This introduces another bookkeeping cost that an in-order layout inherently avoids. So the cache is logically 90% smaller and still physically the same size. Many compression results miss this because they measure on pre-allocated contiguous tensors rather than a paged server. There's another problem. Eviction methods pick which tokens to keep by looking at the attention scores themselves (as expected). But fast attention kernels used in production, like FlashAttention, never save those scores. They compute attention in small pieces and throw the full score grid away as they go, which is also why they're fast. So the exact signal eviction methods need isn't available in memory. The workaround is to fall back to eager attention and build the full matrix, which gives up the speed FlashAttention was there to provide. NVIDIA published a method called TriAttention to solve both these problems. It never needs attention scores. Instead, it scores tokens from the geometry of the model's key and query vectors before RoPE is applied, where those vectors sit in stable clusters. For the memory problem, it runs a compaction pass every 128 decoded tokens. The surviving tokens slide forward to close the holes eviction creates, so whole blocks empty out and return to the allocator while the cache stays in token order. On long reasoning traces, the approach matches full-attention accuracy while decoding 2.5x faster and using 10.7x less KV memory. KV cache compression is a big infrastructure problem. The number that decides whether it works is the count of freed blocks, not the count of evicted tokens. You can find the NVIDIA write-up here: I wrote a first-principles breakdown of how the KV cache works. It walks through why the model stores keys and values at all, why the cache grows with every token, and a comparison of LLM generation speed with and without KV caching. Read it below.
0:23

A tricky LLM interview question: You're serving a reasoning model on vLLM, and it keeps running out of GPU memory on long traces. So you add KV cache compression and evict 90% of the cached tokens. VRAM usage stays as is and GPU still runs out of memory. Why? (answer below) Evicting 90% of the KV cache can free almost none of the memory it was using. This sounds counterintuitive, but it follows directly from how production servers store the cache today. The KV cache grows with every token a model generates. Each token appends its key and value vectors across every layer, and nothing is freed while generation continues. This is the dominant memory cost for reasoning models. If a 32K-token CoT caches ~32K tokens of KV vectors, a Qwen3-32B with 4-bit weights will run out-of-memory around 24K tokens on a 24GB GPU. One obvious solution is to keep the important tokens and drop the rest, since attention is sparse enough to allow it. But this does not solve the memory problem yet. The reason is paged attention, which is the memory manager behind vLLM and most production servers. Under the hood, it splits GPU memory into fixed physical blocks, each one holds the KV for about 16 tokens. This block returns to the allocator only when every slot inside it is empty. Since the eviction logic selects tokens by importance, and such tokens are scattered across blocks... ...so despite eviction, almost every block is left with at least some survivor tokens. For instance, if the logic evicts 14k of 16k tokens across 1,000 blocks, most likely every block will still have a token. This means the allocator frees almost nothing. Placing the new tokens into those freed slots is not ideal because it breaks the cache's layout. Say token 16,001 arrives, and it's placed in the slot the 40th token used to hold. The cache now reads position 38, then 16,001, then 41, so the cache is no longer in token order. Attention can still compute the right answer from that, but only if every slot now carries a separate note recording which position it actually holds. This introduces another bookkeeping cost that an in-order layout inherently avoids. So the cache is logically 90% smaller and still physically the same size. Many compression results miss this because they measure on pre-allocated contiguous tensors rather than a paged server. There's another problem. Eviction methods pick which tokens to keep by looking at the attention scores themselves (as expected). But fast attention kernels used in production, like FlashAttention, never save those scores. They compute attention in small pieces and throw the full score grid away as they go, which is also why they're fast. So the exact signal eviction methods need isn't available in memory. The workaround is to fall back to eager attention and build the full matrix, which gives up the speed FlashAttention was there to provide. NVIDIA published a method called TriAttention to solve both these problems. It never needs attention scores. Instead, it scores tokens from the geometry of the model's key and query vectors before RoPE is applied, where those vectors sit in stable clusters. For the memory problem, it runs a compaction pass every 128 decoded tokens. The surviving tokens slide forward to close the holes eviction creates, so whole blocks empty out and return to the allocator while the cache stays in token order. On long reasoning traces, the approach matches full-attention accuracy while decoding 2.5x faster and using 10.7x less KV memory. KV cache compression is a big infrastructure problem. The number that decides whether it works is the count of freed blocks, not the count of evicted tokens. You can find the NVIDIA write-up here: I wrote a first-principles breakdown of how the KV cache works. It walks through why the model stores keys and values at all, why the cache grows with every token, and a comparison of LLM generation speed with and without KV caching. Read it below.

Avi Chawla

267,206 просмотров • 19 дней назад

HERMES AGENT + OBSIDIAN IS A COMBINATION NOBODY IS TALKING ABOUT. Hermes ships with a bundled Obsidian skill. read, search, and create notes in your vault out of the box. why this combination is powerful: Hermes built-in memory is capped. MEMORY.md: 2,200 chars (~800 tokens). USER.md: 1,375 chars (~500 tokens). Obsidian vault has no cap. your agent writes research, session summaries, project context, and learned patterns as linked markdown notes. unlimited depth. the agent creates indexed notes by design. timestamps, backlinks, tags. every note connects to the knowledge graph. three ways to integrate: 1. BUNDLED OBSIDIAN SKILL (simplest) ships with Hermes. reads, searches, creates notes in your vault directly. hermes skills list | grep obsidian 2. OBSIDIAN MCP SERVER (deepest) 30+ tools: full-text search, tag lookup, note management, vault analysis, link analysis, orphan detection. add it via: hermes mcp 3. TELEGRAM + CRON → VAULT (always-on) set a cron job that writes daily summaries, research findings, or task reports directly into your Obsidian vault. your agent feeds the vault while you sleep. you review in Obsidian when you're ready. the unlock: Hermes memory handles what the agent needs to know per session (capped, injected). Obsidian handles everything the agent has ever learned (uncapped, searchable). short-term in Hermes. long-term in Obsidian. both accessible. both persistent. keep the vault scope narrow at first. start with one /Hermes folder. expand once you trust the workflow. 8 Loops Indise Hermes Agent👇
0:56

HERMES AGENT + OBSIDIAN IS A COMBINATION NOBODY IS TALKING ABOUT. Hermes ships with a bundled Obsidian skill. read, search, and create notes in your vault out of the box. why this combination is powerful: Hermes built-in memory is capped. MEMORY.md: 2,200 chars (~800 tokens). USER.md: 1,375 chars (~500 tokens). Obsidian vault has no cap. your agent writes research, session summaries, project context, and learned patterns as linked markdown notes. unlimited depth. the agent creates indexed notes by design. timestamps, backlinks, tags. every note connects to the knowledge graph. three ways to integrate: 1. BUNDLED OBSIDIAN SKILL (simplest) ships with Hermes. reads, searches, creates notes in your vault directly. hermes skills list | grep obsidian 2. OBSIDIAN MCP SERVER (deepest) 30+ tools: full-text search, tag lookup, note management, vault analysis, link analysis, orphan detection. add it via: hermes mcp 3. TELEGRAM + CRON → VAULT (always-on) set a cron job that writes daily summaries, research findings, or task reports directly into your Obsidian vault. your agent feeds the vault while you sleep. you review in Obsidian when you're ready. the unlock: Hermes memory handles what the agent needs to know per session (capped, injected). Obsidian handles everything the agent has ever learned (uncapped, searchable). short-term in Hermes. long-term in Obsidian. both accessible. both persistent. keep the vault scope narrow at first. start with one /Hermes folder. expand once you trust the workflow. 8 Loops Indise Hermes Agent👇

YanXbt

20,617 просмотров • 1 месяц назад

The modern state highway from Rome to Brindisi still broadly follows the same geographic corridor as a road the ancient Romans began building more than 2,300 years ago. And it's not a coincidence... A 2023 study in the Journal of Regional Science found that modern Italian motorways and railways still largely trace the paths of the old consular roads. The Roman network, the authors write, became "the foundational physical capital" of Italy's current transport system. That network began with a single road. It's called Appian Way. The poet Statius called it regina viarum — the queen of roads. Construction began as a military project during the Samnite Wars to connect Rome to Capua. Over the next century, as Rome pushed south, the road kept extending with it, until it reached Brindisi on the heel of Italy — roughly 540 kilometers of stone. The engineering behind it is extraordinary. Deep foundations of cemented rubble. A surface of polygonal blocks of volcanic basalt, cut and fitted so tightly the historian Procopius later wrote they looked "grown together rather than set by hand." The stretch closest to Rome — the old Appian Way — is a free public park. The original stones are still there — the same ones walked by Roman legions and medieval pilgrims. Cars still use the first few kilometers and after that, the road belongs to whoever wants to walk it. Much of what shaped the ancient world moved along the Appian Way... In 71 BC, after Crassus defeated Spartacus, six thousand of his followers were crucified along a nearly two-hundred-kilometer stretch of it as a warning. The Apostle Paul was brought into Rome as a prisoner on this road. And, according to tradition, Saint Peter walked it the other way, fleeing Nero's persecution. If you want to learn more about one of the oldest and most important of the great ancient roads — and about five other wonders built by the Romans — you can check out today's article here: It's a 5-minute read, and the engineering details alone are worth it. And if you enjoy this kind of deep dive into history, follow and subscribe — there's a new one every week.
0:45

The modern state highway from Rome to Brindisi still broadly follows the same geographic corridor as a road the ancient Romans began building more than 2,300 years ago. And it's not a coincidence... A 2023 study in the Journal of Regional Science found that modern Italian motorways and railways still largely trace the paths of the old consular roads. The Roman network, the authors write, became "the foundational physical capital" of Italy's current transport system. That network began with a single road. It's called Appian Way. The poet Statius called it regina viarum — the queen of roads. Construction began as a military project during the Samnite Wars to connect Rome to Capua. Over the next century, as Rome pushed south, the road kept extending with it, until it reached Brindisi on the heel of Italy — roughly 540 kilometers of stone. The engineering behind it is extraordinary. Deep foundations of cemented rubble. A surface of polygonal blocks of volcanic basalt, cut and fitted so tightly the historian Procopius later wrote they looked "grown together rather than set by hand." The stretch closest to Rome — the old Appian Way — is a free public park. The original stones are still there — the same ones walked by Roman legions and medieval pilgrims. Cars still use the first few kilometers and after that, the road belongs to whoever wants to walk it. Much of what shaped the ancient world moved along the Appian Way... In 71 BC, after Crassus defeated Spartacus, six thousand of his followers were crucified along a nearly two-hundred-kilometer stretch of it as a warning. The Apostle Paul was brought into Rome as a prisoner on this road. And, according to tradition, Saint Peter walked it the other way, fleeing Nero's persecution. If you want to learn more about one of the oldest and most important of the great ancient roads — and about five other wonders built by the Romans — you can check out today's article here: It's a 5-minute read, and the engineering details alone are worth it. And if you enjoy this kind of deep dive into history, follow and subscribe — there's a new one every week.

James Lucas

90,710 просмотров • 2 месяцев назад

🎬 $dTAO Creates The Perfect AI Market EXPLAINED 🌊 Curious about the upcoming Bittensor and $TAO changes? Let's dig into it in this Video. The Core of it 🎯 What Changed: • $TAO emission tied to activity • 7,200 blocks per day • No inflation when demand high • PERFECT BALANCE! Think of it like a smart thermostat: - Too hot? Emissions stop - Too cold? Chain buys - Just right? Natural growth Before, $TAO was emitted at a fixed rate, no matter what. Whether people were staking or not, $TAO was just spilling out. It was kind of like leaving the faucet running all day. Sure, it worked, but it wasn’t efficient or sustainable. Now? It’s different. $TAO emissions are tied to actual demand. Picture a thermostat: If the market’s overheated (lots of $TAO being staked), emissions stop. Why print more $TAO if everyone’s already piling in? If things cool down (not enough $TAO being staked), the chain “buys back” to keep everything balanced. And when it’s just right? The system runs naturally, letting markets do their thing. It’s a self-regulating system. No one’s forcing anything. It’s pure market dynamics Market 🔥 Previous System: - Fixed emissions - Manual controls - Price limitations - ARTIFICIAL LIMITS! New Reality: • Market controls flow • Natural price discovery • Organic growth • PURE ECONOMICS! The old system was like having a boss micro-managing everything kind of like manual controls, price caps, artificial rules. It worked but felt clunky. Now, the market decides: $TAO flows when it’s needed, not when it’s not. Prices? Found naturally, based on real supply and demand. Growth? Pure and organic. This isn’t just a minor upgrade, it’s a complete shift. No more training wheels. This is the economy running on its own. The Number Game 📊 Key Metrics: - 7,200 blocks daily - $1B subnet value(based on year and current prices) - 300 days no inflation - REAL DEMAND! When demand > 7,200 $TAO daily: • Emissions pause • Value increases • Scarcity grows • NATURAL PRESSURE! Here’s how it works, practically speaking: 7,200 blocks/day: That’s how often the chain checks what’s happening. No inflation when more than 7,200 $TAO is staked daily. $1B market cap? That could mean no $TAO emissions for almost a year. Think about it: If demand for $TAO stays high, the system stops flooding the market with new tokens. Instead, it tightens the supply, making $TAO scarcer and more valuable. Less supply = more pressure = higher prices. Growth Engine 🚀 This Creates: 1. Perfect Staking: - Stake more = Less inflation - More value = More growth - More utility = Higher demand 2. Perfect Balance: • Market sets prices • Activity drives value • Growth controls supply This isn’t just about numbers—it’s about balance. dTAO makes everything self-correcting. Too much demand? TAO emissions pause, creating scarcity. Too little demand? The system buys back TAO to stabilize things. It’s like nature. Balance happens automatically, and the system adapts without anyone interfering. The AI Evolution 🧠 For Subnets: - Unlimited growth potential - Natural selection - Value-based competition - PURE INNOVATION! For AI Agents: • Build freely • Compete naturally • Evolve constantly • INFINITE POTENTIAL! Now let’s zoom out. This isn’t just about TAO prices going up or down—it’s about AI evolution: Subnets (specialized networks) can now grow without limits. AI agents (the brains running these networks) compete freely, innovate faster, and constantly improve. It’s survival of the fittest, but in the digital world. The best subnets win, the most useful AI agents thrive, and everyone benefits from the results. No bottlenecks, no gatekeeping, just pure growth. 🎬 Watch Now: YouTube TikTok:
3:30

🎬 $dTAO Creates The Perfect AI Market EXPLAINED 🌊 Curious about the upcoming Bittensor and $TAO changes? Let's dig into it in this Video. The Core of it 🎯 What Changed: • $TAO emission tied to activity • 7,200 blocks per day • No inflation when demand high • PERFECT BALANCE! Think of it like a smart thermostat: - Too hot? Emissions stop - Too cold? Chain buys - Just right? Natural growth Before, $TAO was emitted at a fixed rate, no matter what. Whether people were staking or not, $TAO was just spilling out. It was kind of like leaving the faucet running all day. Sure, it worked, but it wasn’t efficient or sustainable. Now? It’s different. $TAO emissions are tied to actual demand. Picture a thermostat: If the market’s overheated (lots of $TAO being staked), emissions stop. Why print more $TAO if everyone’s already piling in? If things cool down (not enough $TAO being staked), the chain “buys back” to keep everything balanced. And when it’s just right? The system runs naturally, letting markets do their thing. It’s a self-regulating system. No one’s forcing anything. It’s pure market dynamics Market 🔥 Previous System: - Fixed emissions - Manual controls - Price limitations - ARTIFICIAL LIMITS! New Reality: • Market controls flow • Natural price discovery • Organic growth • PURE ECONOMICS! The old system was like having a boss micro-managing everything kind of like manual controls, price caps, artificial rules. It worked but felt clunky. Now, the market decides: $TAO flows when it’s needed, not when it’s not. Prices? Found naturally, based on real supply and demand. Growth? Pure and organic. This isn’t just a minor upgrade, it’s a complete shift. No more training wheels. This is the economy running on its own. The Number Game 📊 Key Metrics: - 7,200 blocks daily - $1B subnet value(based on year and current prices) - 300 days no inflation - REAL DEMAND! When demand > 7,200 $TAO daily: • Emissions pause • Value increases • Scarcity grows • NATURAL PRESSURE! Here’s how it works, practically speaking: 7,200 blocks/day: That’s how often the chain checks what’s happening. No inflation when more than 7,200 $TAO is staked daily. $1B market cap? That could mean no $TAO emissions for almost a year. Think about it: If demand for $TAO stays high, the system stops flooding the market with new tokens. Instead, it tightens the supply, making $TAO scarcer and more valuable. Less supply = more pressure = higher prices. Growth Engine 🚀 This Creates: 1. Perfect Staking: - Stake more = Less inflation - More value = More growth - More utility = Higher demand 2. Perfect Balance: • Market sets prices • Activity drives value • Growth controls supply This isn’t just about numbers—it’s about balance. dTAO makes everything self-correcting. Too much demand? TAO emissions pause, creating scarcity. Too little demand? The system buys back TAO to stabilize things. It’s like nature. Balance happens automatically, and the system adapts without anyone interfering. The AI Evolution 🧠 For Subnets: - Unlimited growth potential - Natural selection - Value-based competition - PURE INNOVATION! For AI Agents: • Build freely • Compete naturally • Evolve constantly • INFINITE POTENTIAL! Now let’s zoom out. This isn’t just about TAO prices going up or down—it’s about AI evolution: Subnets (specialized networks) can now grow without limits. AI agents (the brains running these networks) compete freely, innovate faster, and constantly improve. It’s survival of the fittest, but in the digital world. The best subnets win, the most useful AI agents thrive, and everyone benefits from the results. No bottlenecks, no gatekeeping, just pure growth. 🎬 Watch Now: YouTube TikTok:

Andy ττ

12,534 просмотров • 1 год назад

On this day 56 years ago, three men fell from the sky in a freezing, half-dead spacecraft and landed in the Pacific Ocean. They had been given almost no chance of coming home alive. Six days earlier, astronauts Jim Lovell, Jack Swigert, and Fred Haise launched on Apollo 13, heading for the Moon. Two days into the flight, 200,000 miles from Earth, an oxygen tank exploded and tore a hole in the side of their ship. Within minutes, they were losing oxygen, losing power, and losing heat. There was no plan for this. No one had trained for it. The Moon landing was abandoned. The only question now was whether three men could survive long enough to get home. Their main ship was dying, so they climbed into a smaller attached craft that was only designed to land on the Moon. It was built for two people for two days. They had to make it last four days for three. The temperature inside dropped to 3 degrees Celsius. Water droplets covered every surface. They rationed drinking water to six ounces per man per day, less than a single cup. Jim Lovell lost 14 pounds in four days. Then the air started going bad. The filters that clean carbon dioxide from the air were running out. Without new ones, the crew would suffocate. The spare filters from the main ship were the wrong shape. Square filters. Round slots. Engineers on the ground grabbed the same materials the astronauts had on board, plastic bags, cardboard, duct tape, and a sock, and built a makeshift adapter on a desk. Then they talked the crew through building an identical one while floating in zero gravity, 200,000 miles away. It worked. To get home, they had to swing around the far side of the Moon and fire their engine at the exact right second. Too steep and they would burn up entering Earth’s atmosphere. Too shallow and they would bounce off it and drift into space forever. The entire world stopped. Over 40 million people watched on television. The Pope led prayers from the Vatican. On April 17, 1970, the spacecraft hit the atmosphere. For four minutes, all radio contact went silent. The heat of re-entry surrounds a spacecraft in a layer of superheated gas that blocks all signals. Controllers on the ground called out. Nothing. The silence stretched past the expected time. One minute late. Still nothing. At one minute and 28 seconds past the deadline, a voice broke through. The parachutes opened. The capsule hit the water. All three men were alive. They never reached the Moon. But the mission became the greatest rescue in the history of space travel. It proved that the most dangerous moment in any journey is not the one you prepare for. It is the one nobody saw coming. Jim Lovell never flew in space again. He never walked on the Moon. Years later, when asked if he considered himself unlucky, he said: “I think of the crew of Apollo 1, who died in a fire before they ever left the ground. I think of the crews who never got to fly at all. No, I regard Apollo 13 as a triumph.”
1:36

On this day 56 years ago, three men fell from the sky in a freezing, half-dead spacecraft and landed in the Pacific Ocean. They had been given almost no chance of coming home alive. Six days earlier, astronauts Jim Lovell, Jack Swigert, and Fred Haise launched on Apollo 13, heading for the Moon. Two days into the flight, 200,000 miles from Earth, an oxygen tank exploded and tore a hole in the side of their ship. Within minutes, they were losing oxygen, losing power, and losing heat. There was no plan for this. No one had trained for it. The Moon landing was abandoned. The only question now was whether three men could survive long enough to get home. Their main ship was dying, so they climbed into a smaller attached craft that was only designed to land on the Moon. It was built for two people for two days. They had to make it last four days for three. The temperature inside dropped to 3 degrees Celsius. Water droplets covered every surface. They rationed drinking water to six ounces per man per day, less than a single cup. Jim Lovell lost 14 pounds in four days. Then the air started going bad. The filters that clean carbon dioxide from the air were running out. Without new ones, the crew would suffocate. The spare filters from the main ship were the wrong shape. Square filters. Round slots. Engineers on the ground grabbed the same materials the astronauts had on board, plastic bags, cardboard, duct tape, and a sock, and built a makeshift adapter on a desk. Then they talked the crew through building an identical one while floating in zero gravity, 200,000 miles away. It worked. To get home, they had to swing around the far side of the Moon and fire their engine at the exact right second. Too steep and they would burn up entering Earth’s atmosphere. Too shallow and they would bounce off it and drift into space forever. The entire world stopped. Over 40 million people watched on television. The Pope led prayers from the Vatican. On April 17, 1970, the spacecraft hit the atmosphere. For four minutes, all radio contact went silent. The heat of re-entry surrounds a spacecraft in a layer of superheated gas that blocks all signals. Controllers on the ground called out. Nothing. The silence stretched past the expected time. One minute late. Still nothing. At one minute and 28 seconds past the deadline, a voice broke through. The parachutes opened. The capsule hit the water. All three men were alive. They never reached the Moon. But the mission became the greatest rescue in the history of space travel. It proved that the most dangerous moment in any journey is not the one you prepare for. It is the one nobody saw coming. Jim Lovell never flew in space again. He never walked on the Moon. Years later, when asked if he considered himself unlucky, he said: “I think of the crew of Apollo 1, who died in a fire before they ever left the ground. I think of the crews who never got to fly at all. No, I regard Apollo 13 as a triumph.”

Lemma the Optimist

314,140 просмотров • 3 месяцев назад

This 2.11 minute video shows the real and grave problem affecting CR Mumbai suburban railway. This is Diva Junction station, this afternoon. The level crossing gate here blocks all six lines that means local and NATIONAL lines. This leads to detention of ALL trains on either side of the gate and MASSIVE commuter crowds piling up at all subsequent stations. The road vehicles clearly need to be on a road bridge & not blocking the crucial Mumbai railway like this. And guess what, the road bridge has been built, but the problem are the approaches. Yes. The bridge portion over the railway line is ready, but there is no land for approaches because there are structures, buildings and residents. Residents don't want to move and local landowners have alleged injustice. Was it all not thought before or planned? The Thane Municipal Corporation is now working on this to clear up the mess. Nevertheless, bureaucrats of any standard here have little say in the area since it is a political hotbed of every party. Every decision taken here is a race on political lines. The local public representatives are towering personalities but blinkered by narrow vision. The management at the gate too is poor. This afternoon, the MSF (Maharashtra Security Force) staff had been manning it and when it became unmanageable, RPF (Railway Protection Force) & GRP (Government Railway Police) had to be called (by station announcement) followed by the station staff at the site. The first thing that was done by them was taking pictures probably to prove their presence to the higher ups. The response was not quick but almost idle. It needed to be faster given the stakes (If they understood them.) But its all a mess that needs to be sorted at a different level not with the local cops. They have other priorities. The sad part is that all of this is affecting lakhs of commuters every single day. After the level crossing closed in about 20+ minutes, the first trains that came from both directions were massively crowded with commuters hanging out even in the afternoon non-peak hours. I managed to get in the first train. As it moved ahead to subsequent stations, the crowds got larger & aggressive as they had been waiting for a lonng time. Sadly, they are used to it and manage the lethal commute deftly. Those who can't end up becoming a statistic.
2:11

This 2.11 minute video shows the real and grave problem affecting CR Mumbai suburban railway. This is Diva Junction station, this afternoon. The level crossing gate here blocks all six lines that means local and NATIONAL lines. This leads to detention of ALL trains on either side of the gate and MASSIVE commuter crowds piling up at all subsequent stations. The road vehicles clearly need to be on a road bridge & not blocking the crucial Mumbai railway like this. And guess what, the road bridge has been built, but the problem are the approaches. Yes. The bridge portion over the railway line is ready, but there is no land for approaches because there are structures, buildings and residents. Residents don't want to move and local landowners have alleged injustice. Was it all not thought before or planned? The Thane Municipal Corporation is now working on this to clear up the mess. Nevertheless, bureaucrats of any standard here have little say in the area since it is a political hotbed of every party. Every decision taken here is a race on political lines. The local public representatives are towering personalities but blinkered by narrow vision. The management at the gate too is poor. This afternoon, the MSF (Maharashtra Security Force) staff had been manning it and when it became unmanageable, RPF (Railway Protection Force) & GRP (Government Railway Police) had to be called (by station announcement) followed by the station staff at the site. The first thing that was done by them was taking pictures probably to prove their presence to the higher ups. The response was not quick but almost idle. It needed to be faster given the stakes (If they understood them.) But its all a mess that needs to be sorted at a different level not with the local cops. They have other priorities. The sad part is that all of this is affecting lakhs of commuters every single day. After the level crossing closed in about 20+ minutes, the first trains that came from both directions were massively crowded with commuters hanging out even in the afternoon non-peak hours. I managed to get in the first train. As it moved ahead to subsequent stations, the crowds got larger & aggressive as they had been waiting for a lonng time. Sadly, they are used to it and manage the lethal commute deftly. Those who can't end up becoming a statistic.

Rajendra B. Aklekar

137,004 просмотров • 5 месяцев назад

2001. Larry Page and Sergey Brin sit for their first-ever television interview. Google has 200 employees. They explain that the company almost didn't get off the ground because they couldn't cash a check. The check was for $100,000. It came from Andy Bechtolsheim, one of the co-founders of Sun Microsystems. Page and Brin showed him what they'd built. He said, "This is great, how about I write you a check?" and just wrote it out. Made it out to Google. The problem was that Google didn't exist as a company yet. There was no bank account. No lawyers. No incorporation paperwork. The check sat in Larry Page's desk drawer for a month. They literally could not deposit it. They're both in their late twenties in this interview. They met at Stanford as PhD students and, by their own account, disliked each other from the start. Brin says Page is "kind of obnoxious." Page doesn't disagree. Brin says they argued about everything, debated every single point, and then realized that was their commonality. They became friends, started building a search engine they never planned to build, and put their PhDs on hold to get it out into the world. The part that stings watching this in 2026 is the rejection tour. Before starting Google, they approached existing search companies to sell or license the technology. They went to Yahoo. David Filo, one of Yahoo's founders, told them, "This is great search technology. Why don't you guys make a company, and maybe we'll use you someday?" They went to Excite. They went to InfoSeek. Same response. Page says a CEO at one of those companies told them: "If our search is 85% as good as the next guy's, that's good enough for us." Page and Brin didn't buy that. They thought the search was too important to be 85% as good. So they started Google. No marketing. No ad campaign. They launched it at Stanford, and it grew 20% per month, every single month, for three years straight. Pure word of mouth. By the time of this interview, they're handling over 100 million searches a day. They get 500 resumes in the mail every single day. The office space around them is 30% vacant because the dot-com bubble just popped, but Google is profitable. Page makes a point of this: "We've been really interested in being profitable, like long before it was fashionable." They'd also just hired Eric Schmidt, former CTO of Sun, as CEO. Brin's explanation for why: "Parental supervision, to be honest." Page adds that they're "past the age where we're rebellious" and that running a search engine used by 100 million people a day with 200 employees is "a large responsibility." The number that caught my eye: when Google started in 1998, it indexed 30 million web pages. At the time of this interview, three years later, they indexed 1.3 billion. The page says that if you printed them all out and stacked the paper, it would be about 70 miles high. And it was doubling every year. Every search company they approached turned them down. Yahoo eventually came back and hired Google to power its own search results. The CEO who thought 85% was good enough ran a company that no longer exists. Alphabet, Google's parent company, is worth about $3.6 trillion today. It has about 190,000 employees. That $100,000 check sat in a desk drawer because nobody had incorporated the company. Bechtolsheim's stake from that investment is now worth billions.
14:09

2001. Larry Page and Sergey Brin sit for their first-ever television interview. Google has 200 employees. They explain that the company almost didn't get off the ground because they couldn't cash a check. The check was for $100,000. It came from Andy Bechtolsheim, one of the co-founders of Sun Microsystems. Page and Brin showed him what they'd built. He said, "This is great, how about I write you a check?" and just wrote it out. Made it out to Google. The problem was that Google didn't exist as a company yet. There was no bank account. No lawyers. No incorporation paperwork. The check sat in Larry Page's desk drawer for a month. They literally could not deposit it. They're both in their late twenties in this interview. They met at Stanford as PhD students and, by their own account, disliked each other from the start. Brin says Page is "kind of obnoxious." Page doesn't disagree. Brin says they argued about everything, debated every single point, and then realized that was their commonality. They became friends, started building a search engine they never planned to build, and put their PhDs on hold to get it out into the world. The part that stings watching this in 2026 is the rejection tour. Before starting Google, they approached existing search companies to sell or license the technology. They went to Yahoo. David Filo, one of Yahoo's founders, told them, "This is great search technology. Why don't you guys make a company, and maybe we'll use you someday?" They went to Excite. They went to InfoSeek. Same response. Page says a CEO at one of those companies told them: "If our search is 85% as good as the next guy's, that's good enough for us." Page and Brin didn't buy that. They thought the search was too important to be 85% as good. So they started Google. No marketing. No ad campaign. They launched it at Stanford, and it grew 20% per month, every single month, for three years straight. Pure word of mouth. By the time of this interview, they're handling over 100 million searches a day. They get 500 resumes in the mail every single day. The office space around them is 30% vacant because the dot-com bubble just popped, but Google is profitable. Page makes a point of this: "We've been really interested in being profitable, like long before it was fashionable." They'd also just hired Eric Schmidt, former CTO of Sun, as CEO. Brin's explanation for why: "Parental supervision, to be honest." Page adds that they're "past the age where we're rebellious" and that running a search engine used by 100 million people a day with 200 employees is "a large responsibility." The number that caught my eye: when Google started in 1998, it indexed 30 million web pages. At the time of this interview, three years later, they indexed 1.3 billion. The page says that if you printed them all out and stacked the paper, it would be about 70 miles high. And it was doubling every year. Every search company they approached turned them down. Yahoo eventually came back and hired Google to power its own search results. The CEO who thought 85% was good enough ran a company that no longer exists. Alphabet, Google's parent company, is worth about $3.6 trillion today. It has about 190,000 employees. That $100,000 check sat in a desk drawer because nobody had incorporated the company. Bechtolsheim's stake from that investment is now worth billions.

Anish Moonka

12,042 просмотров • 3 месяцев назад

Tesla is deploying $50 BILLION across 6 factories, a chip fab, robot production lines, AI supercomputers, lithium refineries, and solar manufacturing. To put that in perspective: Tesla made $477 million in profit last quarter. And is investing at roughly 100x that rate. Every other CEO on Earth would get fired for that ratio. Elon's doing it on purpose. Here's what he's assembling: - Own chip factory (TERAFAB with Intel, $25 billion, targeting 1 terawatt of AI compute per year) - Own energy grid (Megapacks powering entire cities) - Own robot workforce (Optimus production starting this year, 1 million units per year at Fremont, 10 million per year planned at Giga Texas) - Own transportation network (robotaxi live in Austin, Dallas, Houston with zero accidents, expanding to 9+ cities) - Own AI training infrastructure (Cortex 2 supercomputer online, 280,000 GPUs by June) - Own lithium refinery (Texas, ramping now) - Own solar panels (new design with 3x the power zones of conventional panels) - Own satellite compute (80% of TERAFAB output going to SpaceX orbital AI satellites) This is just insane. No company in history has attempted to own this many layers of its own supply chain simultaneously. Amazon took 20 years to become profitable because Bezos reinvested every dollar into infrastructure. Wall Street called him insane the entire time. Elon is running the same playbook but across MORE industries, at a FASTER pace, and with technology that didn't exist 5 years ago. The TERAFAB alone is designed to produce 70% of the output of the world's largest semiconductor foundry. Under one roof. Logic chips, memory, and packaging all vertically integrated. But why is he doing this? Elon said existing suppliers including TSMC, Samsung, and Micron simply cannot supply Tesla at the levels it needs. When you can't buy enough of what you need, you build the factory yourself. That's the Henry Ford playbook from 1920. Ford owned the rubber plantations, the iron mines, the glass factories, the railroads, and the forests that supplied his assembly lines. Elon is doing the same thing. Except his version includes orbital data centers, humanoid robots, and autonomous vehicles. The AI5 chip is already taped out. His team worked 6 months straight through holidays and weekends to finish early. He called it the best edge compute inference chip in existence. They're already designing AI6 AND Dojo 3. Meanwhile Tesla's FSD has 1.3 million paid subscribers globally. Record new subscriptions last quarter. Regulatory approval just landed in the Netherlands. China approvals expected by Q3. While every other automaker is trying to figure out how to compete with BYD on price, Elon is building the infrastructure layer that makes the car almost irrelevant. Because if you own the chips, the energy, the robots, the AI, the transportation network, AND the manufacturing... The car is just the interface. The real product is the ecosystem. Elon is spending $50 billion to build a parallel economy that doesn't depend on anyone else's supply chain, anyone else's chips, or anyone else's energy grid. That's closer to being a country than just a company. And whether you love him or hate him, nobody else alive is even attempting this.
1:26

Tesla is deploying $50 BILLION across 6 factories, a chip fab, robot production lines, AI supercomputers, lithium refineries, and solar manufacturing. To put that in perspective: Tesla made $477 million in profit last quarter. And is investing at roughly 100x that rate. Every other CEO on Earth would get fired for that ratio. Elon's doing it on purpose. Here's what he's assembling: - Own chip factory (TERAFAB with Intel, $25 billion, targeting 1 terawatt of AI compute per year) - Own energy grid (Megapacks powering entire cities) - Own robot workforce (Optimus production starting this year, 1 million units per year at Fremont, 10 million per year planned at Giga Texas) - Own transportation network (robotaxi live in Austin, Dallas, Houston with zero accidents, expanding to 9+ cities) - Own AI training infrastructure (Cortex 2 supercomputer online, 280,000 GPUs by June) - Own lithium refinery (Texas, ramping now) - Own solar panels (new design with 3x the power zones of conventional panels) - Own satellite compute (80% of TERAFAB output going to SpaceX orbital AI satellites) This is just insane. No company in history has attempted to own this many layers of its own supply chain simultaneously. Amazon took 20 years to become profitable because Bezos reinvested every dollar into infrastructure. Wall Street called him insane the entire time. Elon is running the same playbook but across MORE industries, at a FASTER pace, and with technology that didn't exist 5 years ago. The TERAFAB alone is designed to produce 70% of the output of the world's largest semiconductor foundry. Under one roof. Logic chips, memory, and packaging all vertically integrated. But why is he doing this? Elon said existing suppliers including TSMC, Samsung, and Micron simply cannot supply Tesla at the levels it needs. When you can't buy enough of what you need, you build the factory yourself. That's the Henry Ford playbook from 1920. Ford owned the rubber plantations, the iron mines, the glass factories, the railroads, and the forests that supplied his assembly lines. Elon is doing the same thing. Except his version includes orbital data centers, humanoid robots, and autonomous vehicles. The AI5 chip is already taped out. His team worked 6 months straight through holidays and weekends to finish early. He called it the best edge compute inference chip in existence. They're already designing AI6 AND Dojo 3. Meanwhile Tesla's FSD has 1.3 million paid subscribers globally. Record new subscriptions last quarter. Regulatory approval just landed in the Netherlands. China approvals expected by Q3. While every other automaker is trying to figure out how to compete with BYD on price, Elon is building the infrastructure layer that makes the car almost irrelevant. Because if you own the chips, the energy, the robots, the AI, the transportation network, AND the manufacturing... The car is just the interface. The real product is the ecosystem. Elon is spending $50 billion to build a parallel economy that doesn't depend on anyone else's supply chain, anyone else's chips, or anyone else's energy grid. That's closer to being a country than just a company. And whether you love him or hate him, nobody else alive is even attempting this.

Ricardo

138,454 просмотров • 2 месяцев назад

This is one of the craziest AI launches of 2026 and it came out of basically nowhere (Save this). A company called Subquadratic just shipped SubQ, and the benchmarks are almost hard to believe. To understand why this is such a big deal, you have to understand the fundamental problem that has defined AI for the last decade. Every large language model in existence is built on transformer architecture, and transformers use a mechanism called standard attention that checks every single word in a sequence against every other word. Double the context length and compute doesn't double, it quadruples, triple it and compute goes up nine times. This quadratic scaling is why frontier models have been stuck at roughly 1 million tokens, why running them at those lengths gets expensive fast, and why the AI labs have essentially been printing money charging you more the longer you need the model to think. The industry has known this problem existed since 2017 but they scaled it anyway. SubQ is built from the ground up to solve it. Instead of processing every possible token relationship, SubQ's sparse attention architecture identifies which relationships actually matter and ignores the rest meaning compute is used where it counts and wasted nowhere else. The result is that compute scales linearly with context length instead of exponentially, and the implications of that one architectural shift are enormous. At 12 million tokens, SubQ reduces attention compute by nearly 1,000x compared to standard frontier models and at 1 million tokens, it runs 52x faster than FlashAttention. And it does all of this while posting frontier level accuracy, scoring 95% on the RULER 128K long-context benchmark versus Claude Opus 4.6's 94.8%, and an 81.8 on SWE-Bench Verified coding tasks, besting Opus 4.6 (80.8) and DeepSeek 4.0 Pro. The cost comparison is where it gets genuinely insane. SubQ runs at under $1.50 per million tokens less than 5% of what Claude Opus charges. On the RULER benchmark, running the test with SubQ cost $8, running the same test with Claude Opus cost $2,600 and that's a 300x cost reduction at equivalent or better accuracy.. Subquadratic launched with $29 million in funding, SubQ is available today for early access via API, and SubQ Code, a coding agent built on the architecture ships alongside it. The transformer has been the unchallenged foundation of every major AI system since 2017. SubQ is the first serious evidence that something structurally better might have just arrived.
1:24

This is one of the craziest AI launches of 2026 and it came out of basically nowhere (Save this). A company called Subquadratic just shipped SubQ, and the benchmarks are almost hard to believe. To understand why this is such a big deal, you have to understand the fundamental problem that has defined AI for the last decade. Every large language model in existence is built on transformer architecture, and transformers use a mechanism called standard attention that checks every single word in a sequence against every other word. Double the context length and compute doesn't double, it quadruples, triple it and compute goes up nine times. This quadratic scaling is why frontier models have been stuck at roughly 1 million tokens, why running them at those lengths gets expensive fast, and why the AI labs have essentially been printing money charging you more the longer you need the model to think. The industry has known this problem existed since 2017 but they scaled it anyway. SubQ is built from the ground up to solve it. Instead of processing every possible token relationship, SubQ's sparse attention architecture identifies which relationships actually matter and ignores the rest meaning compute is used where it counts and wasted nowhere else. The result is that compute scales linearly with context length instead of exponentially, and the implications of that one architectural shift are enormous. At 12 million tokens, SubQ reduces attention compute by nearly 1,000x compared to standard frontier models and at 1 million tokens, it runs 52x faster than FlashAttention. And it does all of this while posting frontier level accuracy, scoring 95% on the RULER 128K long-context benchmark versus Claude Opus 4.6's 94.8%, and an 81.8 on SWE-Bench Verified coding tasks, besting Opus 4.6 (80.8) and DeepSeek 4.0 Pro. The cost comparison is where it gets genuinely insane. SubQ runs at under $1.50 per million tokens less than 5% of what Claude Opus charges. On the RULER benchmark, running the test with SubQ cost $8, running the same test with Claude Opus cost $2,600 and that's a 300x cost reduction at equivalent or better accuracy.. Subquadratic launched with $29 million in funding, SubQ is available today for early access via API, and SubQ Code, a coding agent built on the architecture ships alongside it. The transformer has been the unchallenged foundation of every major AI system since 2017. SubQ is the first serious evidence that something structurally better might have just arrived.

Milk Road AI

278,001 просмотров • 2 месяцев назад

THE STRAIT OF HORMUZ JUST HANDED YOU THE TRADE OF THE DECADE And most investors are looking in completely the wrong direction. Brent crude closed above $103 on Friday. Up nearly 40% since the strikes began on February 28. The Strait of Hormuz is effectively shut down. Insurance companies have canceled war risk coverage. Over 150 ships are stranded. Tanker traffic has collapsed to near zero. The IEA just called it the largest supply disruption in the history of the global oil market. Nearly 20 million barrels per day of crude and product flows have been choked off. The US is scrambling. The IEA coordinated the release of 400 million barrels from strategic reserves, the largest such action ever. Trump ordered emergency insurance for tankers. The Navy was told to begin escort operations. But behind closed doors, Navy officials told tanker executives there's currently NO availability for escorts. And no guarantees there will be. Iran holds the upper hand. And the market knows it. But here's why this matters far beyond the oil price: What we're witnessing is the EMification of America in real time. The US launched strikes in the middle of nuclear negotiations. The executive branch has been attacking central bank independence. Budget deficits are running at levels historically associated with emerging market economies. Erratic policymaking. Massive fiscal deficits. Judicial interference with monetary policy. These are EMERGING MARKET characteristics, and yet the US equity market still carries a premium developed market valuation. That premium is evaporating. Emerging markets returned 33% in 2025. The S&P 500 returned 17%. Almost DOUBLE the outperformance. And 2026 is accelerating the trend. Here's what the consensus is missing: EM macro is BETTER than developed market macro right now. Budget deficits as a percent of GDP? Lower in EM. Debt levels? Lower. Inflation? Lower. Forecasted earnings growth? HIGHER. EM earnings are expected to grow 21% to 29% this year versus 13% to 14% for the U.S. Brazilian equities are trading at roughly 9 times CAPE earnings. About HALF where they traded during the last EM rally in 2018. And the positioning is absurd: US institutional investors have essentially not owned China since Trump 1.0. Most portfolio managers working today weren't even in the business the last time EM led, which was 2001 to 2008. Everyone is out of position. Now layer in commodities: The digital eats the physical. Without copper, silicon, aluminum, and power, there IS no AI. Full stop. And fossil fuels and renewables are rallying AT THE SAME TIME. That tells you the world has a massive power demand problem that isn't going away. Oil above $100. Gold above $4,600. Silver above $85. Copper near all-time highs. The commodity super-cycle is confirming itself in real time. The Iran conflict just poured gasoline on it. Now here's the setup: Emerging market equities, China and Latin America in particular. Commodities across the board. Energy, industrial metals, precious metals. And what to avoid? Long-duration developed market sovereign debt. Overweight positions in the Mag 7, priced for a world where everything goes right and nothing disrupts the AI spending fantasy. Leadership batons in global markets shift in multi-year cycles. The US led from 2009 through 2024. 15 years. Now we're in the early innings of a multi-year rotation into emerging markets and commodities. The flows follow the performance. The performance follows the earnings. And the earnings are now better in EM than in the US. At a fraction of the valuation. With better macro fundamentals. And almost nobody owns it. This is the trade.
2:00

THE STRAIT OF HORMUZ JUST HANDED YOU THE TRADE OF THE DECADE And most investors are looking in completely the wrong direction. Brent crude closed above $103 on Friday. Up nearly 40% since the strikes began on February 28. The Strait of Hormuz is effectively shut down. Insurance companies have canceled war risk coverage. Over 150 ships are stranded. Tanker traffic has collapsed to near zero. The IEA just called it the largest supply disruption in the history of the global oil market. Nearly 20 million barrels per day of crude and product flows have been choked off. The US is scrambling. The IEA coordinated the release of 400 million barrels from strategic reserves, the largest such action ever. Trump ordered emergency insurance for tankers. The Navy was told to begin escort operations. But behind closed doors, Navy officials told tanker executives there's currently NO availability for escorts. And no guarantees there will be. Iran holds the upper hand. And the market knows it. But here's why this matters far beyond the oil price: What we're witnessing is the EMification of America in real time. The US launched strikes in the middle of nuclear negotiations. The executive branch has been attacking central bank independence. Budget deficits are running at levels historically associated with emerging market economies. Erratic policymaking. Massive fiscal deficits. Judicial interference with monetary policy. These are EMERGING MARKET characteristics, and yet the US equity market still carries a premium developed market valuation. That premium is evaporating. Emerging markets returned 33% in 2025. The S&P 500 returned 17%. Almost DOUBLE the outperformance. And 2026 is accelerating the trend. Here's what the consensus is missing: EM macro is BETTER than developed market macro right now. Budget deficits as a percent of GDP? Lower in EM. Debt levels? Lower. Inflation? Lower. Forecasted earnings growth? HIGHER. EM earnings are expected to grow 21% to 29% this year versus 13% to 14% for the U.S. Brazilian equities are trading at roughly 9 times CAPE earnings. About HALF where they traded during the last EM rally in 2018. And the positioning is absurd: US institutional investors have essentially not owned China since Trump 1.0. Most portfolio managers working today weren't even in the business the last time EM led, which was 2001 to 2008. Everyone is out of position. Now layer in commodities: The digital eats the physical. Without copper, silicon, aluminum, and power, there IS no AI. Full stop. And fossil fuels and renewables are rallying AT THE SAME TIME. That tells you the world has a massive power demand problem that isn't going away. Oil above $100. Gold above $4,600. Silver above $85. Copper near all-time highs. The commodity super-cycle is confirming itself in real time. The Iran conflict just poured gasoline on it. Now here's the setup: Emerging market equities, China and Latin America in particular. Commodities across the board. Energy, industrial metals, precious metals. And what to avoid? Long-duration developed market sovereign debt. Overweight positions in the Mag 7, priced for a world where everything goes right and nothing disrupts the AI spending fantasy. Leadership batons in global markets shift in multi-year cycles. The US led from 2009 through 2024. 15 years. Now we're in the early innings of a multi-year rotation into emerging markets and commodities. The flows follow the performance. The performance follows the earnings. And the earnings are now better in EM than in the US. At a fraction of the valuation. With better macro fundamentals. And almost nobody owns it. This is the trade.

George Noble

436,700 просмотров • 4 месяцев назад

** MEGA Parodius Scaling Effects Part 1 ** One of the big challenges with the Parodius Megadrive port is Stage 8's boss - The puffer-fish *Pooyan* with his full screen scaling effect. The goal is to be very close to the arcade (with extras on top ) so I thought lets tackle it head on to see how close we can get. I was also keen to jump into another scaling code rabit hole haha. Pyron pulled out all the stops and got me the source frames and reworked the BG tiles for this test - a big thankyou to him , Vector Orbitex is busy working on Stage 2 tracks so the team is working hard all round on this port. The MD has no sprite / background GFX scaling hardware , however the VDPs Vertical scroll can be updated per scanline to help vertical scaling on backgrounds, but there is a cpu cost to manage all the interupts so thats not free either. With the Horizontal scaling there is no help at all , apart from a semi-friendly packed pixel format for the cpu to work with, its not quite chunky format but better than planar format still for scaling. So its falls back to the 68k CPU to do all of the horizontal expansion which is the largest cpu cost. Basically drawing strips of either 1x, 2x, 3x or 4x wide columns at speed. So we are one week into this Boss's routine and you can see from the below video the horizontal scaling is implented ( vertical will be in the next update ) . We are scaling from 1x to 4x in the video below in 74 steps for testing . The column distributions are always a bit painfull to do - thankfully they are all worked out now. This is the third scaler I have built and the goal was with this one to make it really flexible for use in other projects also, sometimes when you optimise something to the last degree all the flexibility gets taken out of it. Currenty scaling at 12-25 FPS update here, I had some rules against some optimisations which I would use and some I wouldn't , thankfully we are a bit ahead of the Arcades animation frame rate here still and I may yet find optimisations that fit within the scope. We have vertical scaling and sprite spikes to add yet so Im hoping i can find a few more optimisations to offset things when they are implemented also. In a scale frame update we are processing close to 42000 pixels in ram before using DMA to send to VRAM . Using a 41x16 (656 tile scale buffer) - single buffered for now due to its size in VRAM. So thats nearly 21k in tiles ! I had to re-organise ram a bit to support a buffer of that size for the stage. The scaling function is written in 68k assembly , with a little C code handling the Vertical interupt code ( so the game logic can actually run & DMA updates etc ) . The DMA routines are in assembly also and customised for large chunk size ( big blocks of tiles ) which suits the scaler. I had some race conditions to sort out where the cpu was faster than DMA (sending tiles from RAM to VRAM ) and in some cases where it wasn't so it had to be balanced. We may be able to add more detail into the top and bottom of the background yet but its low priority for now until all the other bits are in !! #SGDK #SegaMegadrive #Genesis #Parodius
0:30

** MEGA Parodius Scaling Effects Part 1 ** One of the big challenges with the Parodius Megadrive port is Stage 8's boss - The puffer-fish *Pooyan* with his full screen scaling effect. The goal is to be very close to the arcade (with extras on top ) so I thought lets tackle it head on to see how close we can get. I was also keen to jump into another scaling code rabit hole haha. Pyron pulled out all the stops and got me the source frames and reworked the BG tiles for this test - a big thankyou to him , Vector Orbitex is busy working on Stage 2 tracks so the team is working hard all round on this port. The MD has no sprite / background GFX scaling hardware , however the VDPs Vertical scroll can be updated per scanline to help vertical scaling on backgrounds, but there is a cpu cost to manage all the interupts so thats not free either. With the Horizontal scaling there is no help at all , apart from a semi-friendly packed pixel format for the cpu to work with, its not quite chunky format but better than planar format still for scaling. So its falls back to the 68k CPU to do all of the horizontal expansion which is the largest cpu cost. Basically drawing strips of either 1x, 2x, 3x or 4x wide columns at speed. So we are one week into this Boss's routine and you can see from the below video the horizontal scaling is implented ( vertical will be in the next update ) . We are scaling from 1x to 4x in the video below in 74 steps for testing . The column distributions are always a bit painfull to do - thankfully they are all worked out now. This is the third scaler I have built and the goal was with this one to make it really flexible for use in other projects also, sometimes when you optimise something to the last degree all the flexibility gets taken out of it. Currenty scaling at 12-25 FPS update here, I had some rules against some optimisations which I would use and some I wouldn't , thankfully we are a bit ahead of the Arcades animation frame rate here still and I may yet find optimisations that fit within the scope. We have vertical scaling and sprite spikes to add yet so Im hoping i can find a few more optimisations to offset things when they are implemented also. In a scale frame update we are processing close to 42000 pixels in ram before using DMA to send to VRAM . Using a 41x16 (656 tile scale buffer) - single buffered for now due to its size in VRAM. So thats nearly 21k in tiles ! I had to re-organise ram a bit to support a buffer of that size for the stage. The scaling function is written in 68k assembly , with a little C code handling the Vertical interupt code ( so the game logic can actually run & DMA updates etc ) . The DMA routines are in assembly also and customised for large chunk size ( big blocks of tiles ) which suits the scaler. I had some race conditions to sort out where the cpu was faster than DMA (sending tiles from RAM to VRAM ) and in some cases where it wasn't so it had to be balanced. We may be able to add more detail into the top and bottom of the background yet but its low priority for now until all the other bits are in !! #SGDK #SegaMegadrive #Genesis #Parodius

Shannon Birt

25,545 просмотров • 6 месяцев назад

The multi-leader blockchain endgame: competitive information inclusion as a self-reinforcing mechanism for global price discovery - how we got here, and why Aptos is leading the charge Onchain trading is the killer app In the nine years since the launch of programmable transactions on the Ethereum blockchain, onchain trading has revealed itself as the killer use case for blockchains: onchain listings, volume, and total value locked are all growing with no signs of slowing down, due to the censorship-resistant, permissionless, 24/7/365 qualities afforded by decentralized (DeFi) systems. Monolithic parallelism is key In 2020 Solana was first to market with monolithic, parallel execution (as opposed sharded execution which offers parallelism by partitioning global state into separate information silos), establishing a new design paradigm that raised the bar for throughput and latency: put all of the information in one replicated state machine and make it run as fast as possible. This design produces a single, global hub for activity, liquidity, and token launches, a kind of financial data whiteboard in the sky, where anyone can come and trade at any time with everybody else who has plugged into the system. DEXes are becoming more competitive Historically decentralized systems have been juxtaposed with centralized ones since the latter eliminates the overhead associated with distributed systems coordination. And yet despite this overhead, Solana as a decentralized exchange (DEX) is still pulling in billions of trading volume per day, exceeding that of all but the largest centralized crypto exchanges (CEXs), that simply can't compete with the giant DEX in the sky on token listings or fees. After all, CEXs have to pay for server space, salaries, and lawyers, while a DEX outsources everything. The colocation arms race The one place where CEXs have an advantage over DEXs is on end-to-end latency for colocation applications, or in other words: someone sets up a trading bot in the same data center as the exchange, and their trades get to the exchange faster than everyone else's. When there is only one data ingestion point the fastest trader wins, and after the arms race has played out everyone ends up huddling around the trading hub, effectively cutting off the rest of the world from playing the latency trading game. This is the model that traditional securities exchanges like the Nasdaq or the NYSE 🏛 employ, and because they own the server they can effectively charge whatever they want for access to it. The colocation arms race is also why L2s will probably never decentralize: running the sequencer is practically the same as running the NASDAQ, with the same monopoly on transaction fees collected from a nearby cluster of trading bots (I understand from conversations with Logan Jastremski that the Arbitrum arms race has already hit a Nash Equilibrium in Portland, Oregon). Colocation is a trap But once the colocation arms race has played out, trades become less about incorporating new information in the market and more about skimming off the top by spoofing all of the trades coming in from the other bots. High-frequency trading (HFT) bots located in the NYSE New Jersey data center, for example, are constantly placing buys and sell orders that they have no intention of executing, just to spoof the other colocated bots who are playing the same adversarial game. Information inclusion, on the other hand, the synthesis of real-time world events into prices, takes a back seat because anyone who tries to include new information first needs to batch up their order and send it through a series of middlemen before it ultimately ends up on the exchange: you, I, or practically any other individual can not actually "trade on the NASDAQ", no, we have to express our intent to someone like Robinhood, who then sells our order flow to @CitadelSecurities, who then sends it to the exchange, oh and by the way it doesn't actually even "clear" or "settle" once it "executes" because for whatever reason the whole systems splits these things up and prevents them from happening instantaneously even though it's 2024 and we have computers. Onchain trading cuts out middlemen This whole mess is why we have onchain trading, and why it's starting to win: if you want a mainline to the exchange, without setting up a server, and you want to trade on a news event without getting immediately frontrun by an HFT bot that is sniffing out the trades of every other HFT bot who is easing in batched up order flow on their own terms, then you submit your order to a node in the blockchain and the information gets included in the price upon ingestion. Oh, and by the way the trade is actually fully complete: settled, cleared, reconciled, done, whatever you want to call it, because the people who build decentralized finance (DeFi) build it how it should actually work, not in a way that creates a million incumbents and charges exorbitant rents for access to the system. Onchain trading better for price discovery And the beautiful part about this is that even if a distributed system has more latency than a centralized system, DeFi still ends up incorporating more information into the price faster than centralized finance, because with DeFi the information gets included in the system as soon as it is submitted, not after it has been batched up and sent through a series of middlemen. The consensus mechanism of the blockchain disseminates the information around the world in the form of a price update, while the centralized exchange model requires information about the event to first get propagate to the region of the trading hub, then to get submitted to the colocation server. This means that in terms of global price discovery, onchain trading is strictly a better system because the entire consensus model is based around accelerated information propagation. Because price discovery is a global phenomenon, blockchains, which are global, are actually better than the centralized status quo, on a performance basis, not just from an ideological or convenience-based view. And it has to be multi-leader In practice, effective global information synthesis of information has an additional key requirement: multi-leader architecture. That is, in a single-leader blockchain like Solana, where one validator at a time has a monopoly on ordering transactions into blocks, for their duration as a leader they effectively function as a colocation server. This means that if the current leader is in New York, someone in Singapore who wants to trade on local news as soon as it breaks will still need to get their order all the way around the world to the leader, who is effectively serving as the chain's data ingestion point, before the order can start propagating through the network. But this is issue solved by the introduction of multiple distributed leaders, because then anyone with access to new information can submit their order to the leader closest to them, yielding faster information inclusion in the form of price updates. Multi-leader is also required for fair markets A multi-leader architecture is also required for fair markets, because in a single-leader system the leader has the power to censor transactions, reorder them to their advantage, or even replace transactions with copycats that extract maximum value by replacing the sender's address with their own. For example if someone wants to capture an arbitrage opportunity between two onchain DEXes, they'll need to submit a transaction to the leader and trust that the leader won't simply copy the transaction and submit it themselves. But when there are two or more leaders, users whose transactions are censored by one leader will simply work with a different leader the next time around, eventually cutting off transaction fee flow to the extractive leader. Beyond just strict inclusion, in a multi-leader architecture validators are also forced to compete with each other on latency, because the leader who is fastest at disseminating users' transactions across the network will over time gobble up the largest share of the order flow. Transparent priority fees are a must, or a private mempool will emerge But in order to make this work, a multi-leader architecture must also offer users the ability to pay priority fees AKA "tips" or "bribes" to move their transaction to the front of the line: if there is a $5 arbitrage opportunity onchain, users need to have assurance that they if they pay a 4.99 priority fee to take that arb, they will get priority over a different user who is only willing to tip 4.98. If the native blockchain system does not offer this fair market priority fee mechanism, then it is only a matter of time before one spontaneously emerges in the form of a private mempool like Jito, which can create centralization pressures and undermine the integrity of the system as a whole. Competitive payment for order flow is the stable solution With the right architecture in place, the end result is a competitive environment where endpoints running maximum extractable value (MEV) bots compete with one to offer users the best price for their order flow. In other words, if a user wants to submit an order that can get sandwich attacked for as much as $2 of MEV, then the order should ultimately go to the endpoint bot that is willing to pay the user as much as $1.99 for the right to process their transaction. The price that the provider is willing to pay is ultimately a function of how much in priority fees they might need to pay to the current leader (0 they are the current one), but notably at each stage there is a competitive market for order flow, whether in the form of retail trader's orders, or priority fees among bots that might be forwarding orders to one of the leaders. AptosLabs is already building all this With a public mempool and transaction priority fees, Aptos additionally includes a pipelined architecture that already includes concurrent batching of transactions into blocks, with a single consensus leader who propagates the batched blocks out to the network. And the team is already researching running multiple instances of the consensus algorithm in parallel, yielding multiple consensus leaders who can compete with each other on latency and inclusion - just ask pranav | Shelby, Alexander Spiegelman, and Zekun Li. This means that block times can shrink as the number of consensus leaders grows, with each leader having its own geographical radius of inclusion beyond which it makes more sense to submit to a different leader. The starting point? Something like 60 ms blocks and 3 consensus leaders, partitioning the global information space into competitive and constantly-rotating regions of information inclusion. Messaging is important With concurrent pipelined transaction batching, a public mempool, priority fees, and a clear path to a multi-leader architecture, Aptos leads the industry in onchain trading infrastructure that can truly supplant the centralized colocation paradigm that has heretofore dominated global finance - by offering a truly superior product. And I am hopeful that this deep dive is the first step in communicating not how or that superior product is getting built, but what it means from a bigger picture perspective. If blockchains have found product market fit in anything, it is in trading, and the trading game can only be won by building the biggest, baddest, most high performance system that has as its north star a single, concrete goal: constantly reducing, ever lower toward zero, time time it takes to incorporate information from anywhere in the world into the global price discovery computer. Whoever does this, even 1 ms faster than the competitor, wins the price discovery game, as other blockchains are left in the dust, their DEXes arbed away to zero against the fastest chain on the block. And sure, the blockchain that can rise to this challenge can also handle useful things like payments, NFTs, or other solutions that benefit from permissionlessness and low gas costs, but I want to impress that at the core of this pursuit must be the urge to drive down information inclusion latency to the absolute minimum afforded by the laws of physics through a competitive, market-driven environment. I call on avery.apt 🇺🇸 , CTO of Aptos Labs, to lean in on this messaging, to make it clear that Aptos is here for this singular mission, to build the most performant price discovery engine in history, as a rallying call for alignment in development efforts across the ecosystem and broader industry. Where does this go? As the latencies drop, the spreads tighten, and the information inclusion increases with every incremental increase in network bandwidth, we can expect a new class of competing techno-financial hubs that aggregate around the world's largest information sources: New York, Washington DC, London, Tokyo, etc., commanding stake distribution commensurate with the density of information flow in these respective locales. With the right incentives in place, competing concurrent leaders will invest ever more in infrastructure to get their packets out to the network faster than the rest, yielding clusters of fiber optic cable around the world's financial hubs, neurons in the global financial brain connecting not just HFT firms to servers in their city, but connecting every city with every other city, to move pricing information across oceans and continents. And retail traders, who have been left out of the colocation game, will only benefit: this entire system gets faster, more inclusive, with tighter spreads and lower fees, and it is such an amazing opportunity to watch all of this unfold in real time. The future of blockchains is the future of trading, is the future of competitive information inclusion in real-time, is the future of truly unified global markets, because at the the core of this industry is a simple idea: connect the computers, and see where the incentives lead. They lead to this, and Aptos is leading the charge, because its tech is purpose-built for this exact purpose. So tell the world about it.
2:41

The multi-leader blockchain endgame: competitive information inclusion as a self-reinforcing mechanism for global price discovery - how we got here, and why Aptos is leading the charge Onchain trading is the killer app In the nine years since the launch of programmable transactions on the Ethereum blockchain, onchain trading has revealed itself as the killer use case for blockchains: onchain listings, volume, and total value locked are all growing with no signs of slowing down, due to the censorship-resistant, permissionless, 24/7/365 qualities afforded by decentralized (DeFi) systems. Monolithic parallelism is key In 2020 Solana was first to market with monolithic, parallel execution (as opposed sharded execution which offers parallelism by partitioning global state into separate information silos), establishing a new design paradigm that raised the bar for throughput and latency: put all of the information in one replicated state machine and make it run as fast as possible. This design produces a single, global hub for activity, liquidity, and token launches, a kind of financial data whiteboard in the sky, where anyone can come and trade at any time with everybody else who has plugged into the system. DEXes are becoming more competitive Historically decentralized systems have been juxtaposed with centralized ones since the latter eliminates the overhead associated with distributed systems coordination. And yet despite this overhead, Solana as a decentralized exchange (DEX) is still pulling in billions of trading volume per day, exceeding that of all but the largest centralized crypto exchanges (CEXs), that simply can't compete with the giant DEX in the sky on token listings or fees. After all, CEXs have to pay for server space, salaries, and lawyers, while a DEX outsources everything. The colocation arms race The one place where CEXs have an advantage over DEXs is on end-to-end latency for colocation applications, or in other words: someone sets up a trading bot in the same data center as the exchange, and their trades get to the exchange faster than everyone else's. When there is only one data ingestion point the fastest trader wins, and after the arms race has played out everyone ends up huddling around the trading hub, effectively cutting off the rest of the world from playing the latency trading game. This is the model that traditional securities exchanges like the Nasdaq or the NYSE 🏛 employ, and because they own the server they can effectively charge whatever they want for access to it. The colocation arms race is also why L2s will probably never decentralize: running the sequencer is practically the same as running the NASDAQ, with the same monopoly on transaction fees collected from a nearby cluster of trading bots (I understand from conversations with Logan Jastremski that the Arbitrum arms race has already hit a Nash Equilibrium in Portland, Oregon). Colocation is a trap But once the colocation arms race has played out, trades become less about incorporating new information in the market and more about skimming off the top by spoofing all of the trades coming in from the other bots. High-frequency trading (HFT) bots located in the NYSE New Jersey data center, for example, are constantly placing buys and sell orders that they have no intention of executing, just to spoof the other colocated bots who are playing the same adversarial game. Information inclusion, on the other hand, the synthesis of real-time world events into prices, takes a back seat because anyone who tries to include new information first needs to batch up their order and send it through a series of middlemen before it ultimately ends up on the exchange: you, I, or practically any other individual can not actually "trade on the NASDAQ", no, we have to express our intent to someone like Robinhood, who then sells our order flow to @CitadelSecurities, who then sends it to the exchange, oh and by the way it doesn't actually even "clear" or "settle" once it "executes" because for whatever reason the whole systems splits these things up and prevents them from happening instantaneously even though it's 2024 and we have computers. Onchain trading cuts out middlemen This whole mess is why we have onchain trading, and why it's starting to win: if you want a mainline to the exchange, without setting up a server, and you want to trade on a news event without getting immediately frontrun by an HFT bot that is sniffing out the trades of every other HFT bot who is easing in batched up order flow on their own terms, then you submit your order to a node in the blockchain and the information gets included in the price upon ingestion. Oh, and by the way the trade is actually fully complete: settled, cleared, reconciled, done, whatever you want to call it, because the people who build decentralized finance (DeFi) build it how it should actually work, not in a way that creates a million incumbents and charges exorbitant rents for access to the system. Onchain trading better for price discovery And the beautiful part about this is that even if a distributed system has more latency than a centralized system, DeFi still ends up incorporating more information into the price faster than centralized finance, because with DeFi the information gets included in the system as soon as it is submitted, not after it has been batched up and sent through a series of middlemen. The consensus mechanism of the blockchain disseminates the information around the world in the form of a price update, while the centralized exchange model requires information about the event to first get propagate to the region of the trading hub, then to get submitted to the colocation server. This means that in terms of global price discovery, onchain trading is strictly a better system because the entire consensus model is based around accelerated information propagation. Because price discovery is a global phenomenon, blockchains, which are global, are actually better than the centralized status quo, on a performance basis, not just from an ideological or convenience-based view. And it has to be multi-leader In practice, effective global information synthesis of information has an additional key requirement: multi-leader architecture. That is, in a single-leader blockchain like Solana, where one validator at a time has a monopoly on ordering transactions into blocks, for their duration as a leader they effectively function as a colocation server. This means that if the current leader is in New York, someone in Singapore who wants to trade on local news as soon as it breaks will still need to get their order all the way around the world to the leader, who is effectively serving as the chain's data ingestion point, before the order can start propagating through the network. But this is issue solved by the introduction of multiple distributed leaders, because then anyone with access to new information can submit their order to the leader closest to them, yielding faster information inclusion in the form of price updates. Multi-leader is also required for fair markets A multi-leader architecture is also required for fair markets, because in a single-leader system the leader has the power to censor transactions, reorder them to their advantage, or even replace transactions with copycats that extract maximum value by replacing the sender's address with their own. For example if someone wants to capture an arbitrage opportunity between two onchain DEXes, they'll need to submit a transaction to the leader and trust that the leader won't simply copy the transaction and submit it themselves. But when there are two or more leaders, users whose transactions are censored by one leader will simply work with a different leader the next time around, eventually cutting off transaction fee flow to the extractive leader. Beyond just strict inclusion, in a multi-leader architecture validators are also forced to compete with each other on latency, because the leader who is fastest at disseminating users' transactions across the network will over time gobble up the largest share of the order flow. Transparent priority fees are a must, or a private mempool will emerge But in order to make this work, a multi-leader architecture must also offer users the ability to pay priority fees AKA "tips" or "bribes" to move their transaction to the front of the line: if there is a $5 arbitrage opportunity onchain, users need to have assurance that they if they pay a 4.99 priority fee to take that arb, they will get priority over a different user who is only willing to tip 4.98. If the native blockchain system does not offer this fair market priority fee mechanism, then it is only a matter of time before one spontaneously emerges in the form of a private mempool like Jito, which can create centralization pressures and undermine the integrity of the system as a whole. Competitive payment for order flow is the stable solution With the right architecture in place, the end result is a competitive environment where endpoints running maximum extractable value (MEV) bots compete with one to offer users the best price for their order flow. In other words, if a user wants to submit an order that can get sandwich attacked for as much as $2 of MEV, then the order should ultimately go to the endpoint bot that is willing to pay the user as much as $1.99 for the right to process their transaction. The price that the provider is willing to pay is ultimately a function of how much in priority fees they might need to pay to the current leader (0 they are the current one), but notably at each stage there is a competitive market for order flow, whether in the form of retail trader's orders, or priority fees among bots that might be forwarding orders to one of the leaders. AptosLabs is already building all this With a public mempool and transaction priority fees, Aptos additionally includes a pipelined architecture that already includes concurrent batching of transactions into blocks, with a single consensus leader who propagates the batched blocks out to the network. And the team is already researching running multiple instances of the consensus algorithm in parallel, yielding multiple consensus leaders who can compete with each other on latency and inclusion - just ask pranav | Shelby, Alexander Spiegelman, and Zekun Li. This means that block times can shrink as the number of consensus leaders grows, with each leader having its own geographical radius of inclusion beyond which it makes more sense to submit to a different leader. The starting point? Something like 60 ms blocks and 3 consensus leaders, partitioning the global information space into competitive and constantly-rotating regions of information inclusion. Messaging is important With concurrent pipelined transaction batching, a public mempool, priority fees, and a clear path to a multi-leader architecture, Aptos leads the industry in onchain trading infrastructure that can truly supplant the centralized colocation paradigm that has heretofore dominated global finance - by offering a truly superior product. And I am hopeful that this deep dive is the first step in communicating not how or that superior product is getting built, but what it means from a bigger picture perspective. If blockchains have found product market fit in anything, it is in trading, and the trading game can only be won by building the biggest, baddest, most high performance system that has as its north star a single, concrete goal: constantly reducing, ever lower toward zero, time time it takes to incorporate information from anywhere in the world into the global price discovery computer. Whoever does this, even 1 ms faster than the competitor, wins the price discovery game, as other blockchains are left in the dust, their DEXes arbed away to zero against the fastest chain on the block. And sure, the blockchain that can rise to this challenge can also handle useful things like payments, NFTs, or other solutions that benefit from permissionlessness and low gas costs, but I want to impress that at the core of this pursuit must be the urge to drive down information inclusion latency to the absolute minimum afforded by the laws of physics through a competitive, market-driven environment. I call on avery.apt 🇺🇸 , CTO of Aptos Labs, to lean in on this messaging, to make it clear that Aptos is here for this singular mission, to build the most performant price discovery engine in history, as a rallying call for alignment in development efforts across the ecosystem and broader industry. Where does this go? As the latencies drop, the spreads tighten, and the information inclusion increases with every incremental increase in network bandwidth, we can expect a new class of competing techno-financial hubs that aggregate around the world's largest information sources: New York, Washington DC, London, Tokyo, etc., commanding stake distribution commensurate with the density of information flow in these respective locales. With the right incentives in place, competing concurrent leaders will invest ever more in infrastructure to get their packets out to the network faster than the rest, yielding clusters of fiber optic cable around the world's financial hubs, neurons in the global financial brain connecting not just HFT firms to servers in their city, but connecting every city with every other city, to move pricing information across oceans and continents. And retail traders, who have been left out of the colocation game, will only benefit: this entire system gets faster, more inclusive, with tighter spreads and lower fees, and it is such an amazing opportunity to watch all of this unfold in real time. The future of blockchains is the future of trading, is the future of competitive information inclusion in real-time, is the future of truly unified global markets, because at the the core of this industry is a simple idea: connect the computers, and see where the incentives lead. They lead to this, and Aptos is leading the charge, because its tech is purpose-built for this exact purpose. So tell the world about it.

Alex Kahn

24,432 просмотров • 1 год назад

$EOSE I get it, it’s easy to feel let down when you’re waiting for sales numbers or partnerships to light up the stock, but you get a new product announcement you didn't expect. But hear me out: This marks an inflection point. Don't underestimate this breakthrough innovation. Super bullish on where this takes Eos. This isn’t just another tweak; it’s a game-changer in how energy storage/BESS will be thought about. Sometimes the biggest innovations boil down to the form factor, how you package and deploy the tech. This feels like one of those moments. Look, we’ve all seen the Z3 modules and DawnOS in action. They’re solid: zinc-based chemistry that’s non-flammable, hits 90%+ round-trip efficiency, and holds 96% capacity over 25 years, even in brutal conditions. But up until now, Eos was cramming them into standard containers, trying to fit into what the market expects from BESS. This was never going to work. Indensity flips that script. It’s basically Lego blocks for energy storage. Stackable Indensity Core units that slot into a simple steel frame. You build the superstructure first, then forklift the cores into place. No more suspending modules from container walls; they sit flat on the floor. It’s modular, weather-proof, and rated for indoor or outdoor use. Spread them out in rural spots, stack them high in suburbs to dodge zoning headaches, or go vertical in tight urban areas where land is gold. The breakthrough here? Way more power in the same footprint, powered by the same Z3 modules we know, but packaged smarter. You may think it’s “just” a repackaging, but form factor shifts have crushed it in other industries many times before. Think shipping containers: standardize the box, and suddenly global trade explodes. Even Lego itself started as simple bricks, but the interlocking design created endless possibilities and dominated toys. Indensity does that for BESS, turns rigid containers into flexible stacks that adapt to any site, any scale. This has a huge impact on manufacturing and costs too. On the factory side, this setup screams automation. Their lines can now churn out Z3 modules straight into these steel-framed packs, even pre-stacked in multiple layers. No more wrestling with full containers during assembly. It’s like palletizing goods for a warehouse. Defect rates drop because handling is simpler, and you can extend the production line with a substation that outputs ready-to-ship units. Transportation? Huge win. These cores fit like pallets in standard trucks. No cranes at the destination. Just forklift them off the lorry and slot them in. Installation gets stupid easy too: forklifts instead of heavy cranes mean faster setup, less site prep, and lower labor costs. And with the density jump, you’re using less land overall, which slashes permitting time and real estate expenses in crowded areas. Bottom line: this isn’t competing in the traditional "container BESS market" anymore. It’s carving out a whole new space where no one’s playing yet: hyper-dense, hyper-flexible storage for tough spots like data centers or urban grids. Sure, sodium-ion might show up eventually, but right now? Zero real competition. If Eos nails execution (and yeah, they’ve been busy scaling sales alongside this), investors will wake up to the massive market here. No more apples-to-apples with Lithium-Ion. Mark my words: this is zero-to-one moment for Eos. Well done 🔋Eos Energy Enterprises, Inc.🔋. Hats off. What do you all think?
11:23

$EOSE I get it, it’s easy to feel let down when you’re waiting for sales numbers or partnerships to light up the stock, but you get a new product announcement you didn't expect. But hear me out: This marks an inflection point. Don't underestimate this breakthrough innovation. Super bullish on where this takes Eos. This isn’t just another tweak; it’s a game-changer in how energy storage/BESS will be thought about. Sometimes the biggest innovations boil down to the form factor, how you package and deploy the tech. This feels like one of those moments. Look, we’ve all seen the Z3 modules and DawnOS in action. They’re solid: zinc-based chemistry that’s non-flammable, hits 90%+ round-trip efficiency, and holds 96% capacity over 25 years, even in brutal conditions. But up until now, Eos was cramming them into standard containers, trying to fit into what the market expects from BESS. This was never going to work. Indensity flips that script. It’s basically Lego blocks for energy storage. Stackable Indensity Core units that slot into a simple steel frame. You build the superstructure first, then forklift the cores into place. No more suspending modules from container walls; they sit flat on the floor. It’s modular, weather-proof, and rated for indoor or outdoor use. Spread them out in rural spots, stack them high in suburbs to dodge zoning headaches, or go vertical in tight urban areas where land is gold. The breakthrough here? Way more power in the same footprint, powered by the same Z3 modules we know, but packaged smarter. You may think it’s “just” a repackaging, but form factor shifts have crushed it in other industries many times before. Think shipping containers: standardize the box, and suddenly global trade explodes. Even Lego itself started as simple bricks, but the interlocking design created endless possibilities and dominated toys. Indensity does that for BESS, turns rigid containers into flexible stacks that adapt to any site, any scale. This has a huge impact on manufacturing and costs too. On the factory side, this setup screams automation. Their lines can now churn out Z3 modules straight into these steel-framed packs, even pre-stacked in multiple layers. No more wrestling with full containers during assembly. It’s like palletizing goods for a warehouse. Defect rates drop because handling is simpler, and you can extend the production line with a substation that outputs ready-to-ship units. Transportation? Huge win. These cores fit like pallets in standard trucks. No cranes at the destination. Just forklift them off the lorry and slot them in. Installation gets stupid easy too: forklifts instead of heavy cranes mean faster setup, less site prep, and lower labor costs. And with the density jump, you’re using less land overall, which slashes permitting time and real estate expenses in crowded areas. Bottom line: this isn’t competing in the traditional "container BESS market" anymore. It’s carving out a whole new space where no one’s playing yet: hyper-dense, hyper-flexible storage for tough spots like data centers or urban grids. Sure, sodium-ion might show up eventually, but right now? Zero real competition. If Eos nails execution (and yeah, they’ve been busy scaling sales alongside this), investors will wake up to the massive market here. No more apples-to-apples with Lithium-Ion. Mark my words: this is zero-to-one moment for Eos. Well done 🔋Eos Energy Enterprises, Inc.🔋. Hats off. What do you all think?

Michal Brojak

31,120 просмотров • 6 месяцев назад

I’ve been using GPT-5.6 Sol internally for the past two months, I've spent probably 25+ billion tokens. Here’s my review and comparison to Fable 5: > Let's start with the analogy because everyone seems to be giving theirs - GPT-5.6 is likely the last version of the GPT-5 training run series. It's kind of like an athlete at their peak. Through years of experience in the game, they've become the most reliable player and has the highest game IQ. But, there's no more room to grow. Fable on the other hand, being essentially the first version of a new training run, is the first round draft pick rookie. Raw talent mixed with the energy only a young person would have results in some incredible plays we didn't think possible, but also mistakes due to lack of experience. But that rookie will only improve and likely will be better than the veteran ever was because it's a new game and a new era. > GPT-5.6 is genuinely better at long, sustained work. With /goal, I've had it running complex projects for days with almost no intervention. It built a Minecraft-style game, kept adding features and mobs after the core game worked, and only stopped because I stopped the run. I never felt as though I had to jump in and guide it back to the right path. > It keeps finding useful work when you give it a concrete finish line. I had it recreate Excel with a loop. It inspected the real desktop excel app with Computer Use, comparing that against its own build, and closing the gaps. I stopped it after six days after it had built an incredible amount of functionality. > It's faster than other models in two different ways. The raw generation speed is higher, something OpenAI has been putting effort into. But it also takes a shorter path to solutions. It wanders less, changes less code, and generally knows how to get things done directly. In daily use, it feels about 2-3x times faster than Fable. That's my impression, not a controlled benchmark. The difference is large enough that I notice it constantly. > It works well across a wide range of tasks. I use it for one-line edits, quick questions, browser chores, and multi-day builds without changing my prompting style. Speaking of browser control, its the best ever I've used. To the point where I actually use it often. If a task lives on a website, GPT-5.6 usually opens the browser and does it there instead of asking for an API key or forcing everything through the terminal. When I switched back to GPT-5.5, it went straight to the command line even when the browser was clearly the better tool. > And it can handle real browser work, not just toy demos. During a data import, I had it monitor Supabase and resize instances as the load changed. It stayed on the dashboard, adjusted capacity, and checked the result without an API or a custom script. > I also gave it a full Google Workspace migration. It moved Forward Future from to preserved the old aliases, and configured MX, SPF, and DKIM. Before a consequential save, it stopped, explained exactly what would change, and waited for confirmation. > The reasoning setting matters a lot. Light is good for questions and small edits. High and Extra High are the sweet spots for serious work. Ultra usually takes longer than the extra thinking is worth and burns tokens. > I love that 5.6 is split into 3 sizes. Not only can you control speed and cost that way, but you still also have the thinking effort setting for each of them. Very precise controls. I just wish Codex automatically routed my prompts for me. > Its personality is blunt and a little bland. Claude feels warmer and more natural to talk to. GPT-5.6 is more clinical, but I like that for work. It gives me enough explanation and rarely pads the answer. I usually have to ask Fable to explain things more simply and/or more concise. > Its front-end taste has improved, but the default is predictable. Left alone, it turns websites into PowerPoint decks with huge statements and hard section breaks. The good news is that it takes design direction well and can revise without destroying the parts that already work. > It still makes confident mistakes. I asked it to rebuild parts of a system, and it told me the job was finished. Later, I found out it wasn't. Bits of its internal process also leak into the answer occasionally. > Claude Fable is more naturally autonomous on large, open-ended projects. GPT-5.6 is easier to reach for. I don't need to invent a huge project to justify using it. It works just as well for a small edit or browser chore. > GPT-5.6 is also cheaper. Sol costs $5 per million input tokens and $30 per million output tokens. Fable costs $10 and $50. Cached input is cheaper too. Still, cost per finished task matters more than cost per token. > GPT-5.6 isn't the best at everything, and it still needs supervision. But it generates faster, wanders less, works at almost any scale, and wastes less of my time. It's the model I have the most confidence in to get the job done right the first time. I put together a full breakdown with all the tests, prompts, and examples on a site. You can read it here:
7:46

I’ve been using GPT-5.6 Sol internally for the past two months, I've spent probably 25+ billion tokens. Here’s my review and comparison to Fable 5: > Let's start with the analogy because everyone seems to be giving theirs - GPT-5.6 is likely the last version of the GPT-5 training run series. It's kind of like an athlete at their peak. Through years of experience in the game, they've become the most reliable player and has the highest game IQ. But, there's no more room to grow. Fable on the other hand, being essentially the first version of a new training run, is the first round draft pick rookie. Raw talent mixed with the energy only a young person would have results in some incredible plays we didn't think possible, but also mistakes due to lack of experience. But that rookie will only improve and likely will be better than the veteran ever was because it's a new game and a new era. > GPT-5.6 is genuinely better at long, sustained work. With /goal, I've had it running complex projects for days with almost no intervention. It built a Minecraft-style game, kept adding features and mobs after the core game worked, and only stopped because I stopped the run. I never felt as though I had to jump in and guide it back to the right path. > It keeps finding useful work when you give it a concrete finish line. I had it recreate Excel with a loop. It inspected the real desktop excel app with Computer Use, comparing that against its own build, and closing the gaps. I stopped it after six days after it had built an incredible amount of functionality. > It's faster than other models in two different ways. The raw generation speed is higher, something OpenAI has been putting effort into. But it also takes a shorter path to solutions. It wanders less, changes less code, and generally knows how to get things done directly. In daily use, it feels about 2-3x times faster than Fable. That's my impression, not a controlled benchmark. The difference is large enough that I notice it constantly. > It works well across a wide range of tasks. I use it for one-line edits, quick questions, browser chores, and multi-day builds without changing my prompting style. Speaking of browser control, its the best ever I've used. To the point where I actually use it often. If a task lives on a website, GPT-5.6 usually opens the browser and does it there instead of asking for an API key or forcing everything through the terminal. When I switched back to GPT-5.5, it went straight to the command line even when the browser was clearly the better tool. > And it can handle real browser work, not just toy demos. During a data import, I had it monitor Supabase and resize instances as the load changed. It stayed on the dashboard, adjusted capacity, and checked the result without an API or a custom script. > I also gave it a full Google Workspace migration. It moved Forward Future from to preserved the old aliases, and configured MX, SPF, and DKIM. Before a consequential save, it stopped, explained exactly what would change, and waited for confirmation. > The reasoning setting matters a lot. Light is good for questions and small edits. High and Extra High are the sweet spots for serious work. Ultra usually takes longer than the extra thinking is worth and burns tokens. > I love that 5.6 is split into 3 sizes. Not only can you control speed and cost that way, but you still also have the thinking effort setting for each of them. Very precise controls. I just wish Codex automatically routed my prompts for me. > Its personality is blunt and a little bland. Claude feels warmer and more natural to talk to. GPT-5.6 is more clinical, but I like that for work. It gives me enough explanation and rarely pads the answer. I usually have to ask Fable to explain things more simply and/or more concise. > Its front-end taste has improved, but the default is predictable. Left alone, it turns websites into PowerPoint decks with huge statements and hard section breaks. The good news is that it takes design direction well and can revise without destroying the parts that already work. > It still makes confident mistakes. I asked it to rebuild parts of a system, and it told me the job was finished. Later, I found out it wasn't. Bits of its internal process also leak into the answer occasionally. > Claude Fable is more naturally autonomous on large, open-ended projects. GPT-5.6 is easier to reach for. I don't need to invent a huge project to justify using it. It works just as well for a small edit or browser chore. > GPT-5.6 is also cheaper. Sol costs $5 per million input tokens and $30 per million output tokens. Fable costs $10 and $50. Cached input is cheaper too. Still, cost per finished task matters more than cost per token. > GPT-5.6 isn't the best at everything, and it still needs supervision. But it generates faster, wanders less, works at almost any scale, and wastes less of my time. It's the model I have the most confidence in to get the job done right the first time. I put together a full breakdown with all the tests, prompts, and examples on a site. You can read it here:

Matthew Berman

183,716 просмотров • 7 дней назад

Shabana Mahmood wants you to think it's stopped. It hasn't. They didn't LOSE control of your housing. They didn't run OUT of homes for you. They GAVE them away. She banned the bit you could SEE, and kept the engine running. 21 brand new houses in a quiet Shropshire village. Worth £250,000 each. Built for local families. Handed to the asylum system instead, for up to 83 migrants, rent free. Rural. Back gardens. Next to where local children play. On a street where a British family had already bought and moved in. Residents weren't even told who was coming. They had to FOI their own council for the basics. How many. Who. Whether they're even families. Made to force it out in writing. Asked on live radio why our own homeless get nothing while this happens, Yvette Cooper didn’t deny it. She admitted it, said they should be in more appropriate accommodation. So she knows it’s wrong. Then she swerved. Reduce the system. Family reunion. Student visas. The system we inherited. We're saving a billion. Everything except the answer. So here's the answer she wouldn't give. Who decides it. Not your council. Not your MP. Not one person in that village. The Home Office decides, and tells a private firm, Serco, to place people. We know because Serco said it. The Home Office decides where people go, and instructs us accordingly. The people you vote for have no say. The people with the say, you never voted for. Why they pack them in. They're paid by the head. Twelve hundred pounds a year per person, a hundred a month per extra bed. Six in a house beats a family of four. That's how you get 83 people into 21 homes. Not compassion. Arithmetic. Who even chooses who comes. Under the new scheme, charities, universities and employers pick. Not you. Not your MP. Private groups hold the keys. Who gets rich. The man who owns one of these firms went from a caravan park to a billionaire. Porsches. Mayfair. Monaco. All from housing migrants on our money. His firm paid out twenty eight million in dividends in one year, almost all to him. One man, drawing more from our foreign aid budget than the whole of Ghana. He made so much his firm agreed to hand thirty two million back for breaching the profit cap. Money the Home Office still hasn't confirmed it's even collected. Here’s why they'll NEVER really stop it. The money we spend housing migrants here is counted as foreign aid. On paper, Britain feeding the world's poor. Really, filling a landlord's pockets in Monaco. Because it counts as aid, the more come, the better their books look. Count the ways they win. The contractor gets rich. The Treasury books four point three billion as aid that never leaves the country. Real aid to Africa is cut, the money stays here in Clearsprings and the consultancies, and you pay twice, in tax and in a home your kids will never get. The boats aren't a crisis to them. They're the business model. Forty thousand to leave, called a saving. Ten thousand to stay, called fairness, though their own figures say most will never earn enough to pay a penny. Now, ban the new-builds, and call it fixed. A different headline every month. The same engine underneath. Here’s the smoke, right on cue. They've said Stoke Heath won't go ahead. Sounds like a win. It isn't. They didn't stop it because it was wrong. They stopped it because it was SEEN. It went viral. It got too hot. So they killed the one estate the camera found, and kept every one it didn't. Its not just Shropshire. Over in Suffolk, migrants have already moved into four brand new townhouses worth £300,000, with en-suites, underfloor heating and electric car points, handed over rent free Flat blocks taken in Huddersfield, Chelmsford, Bournemouth. Same Home Office. Same contractor. Town after town, the ones no camera found. They won't even confirm the family already in Stoke Heath will be moved out. The ban only touches new-builds. The flat blocks stay. The HMOs stay. The contracts stay. Who really runs it. The Home Office behind all this was headed by Antonia Romeo, now Cabinet Secretary, the most powerful official in the country. She began her own career at a global consultancy, Oliver Wyman, a firm with a documented history of work in the refugee sector. Nobody voted for her. She was there before this government, and she'll be there after it. Ministers are the face. She's the engine. Three firms. Fifteen billion pounds of contracts. Break clauses that could end it tomorrow. Not one pulled. Yvette Cooper ran this and dodged it on air. Shabana Mahmood bans the one part that got caught and keeps the rest. Same engine. Two faces. The next face is already here. Whoever walks into Number 10 next can't break it either. Romeo stays. The contracts stay. They inherit it intact. Our own keep waiting. 1.3 million on the longest social housing queue in over a decade. The family that saved ten years. The homeless man born here. Passed over, in their own country, so the engine keeps running. You can respray a car. New paint, new bodywork, new badge. It still drives, because you never touched the engine. Same here. Change the government, the face, the party. The engine never moves. The Home Office. The contractor. The billionaire. The consultancy. The money. The one they never let you vote out. That's the engine and every few months they hand you a new face to be angry at, so you keep swinging at the paint. They didn't ban the machine. They banned the photo of it. Then stood at a podium and told us they'd fixed it. That's not a mistake. That's a decision. Stop aiming at the paint. Go for the engine.
1:19

Shabana Mahmood wants you to think it's stopped. It hasn't. They didn't LOSE control of your housing. They didn't run OUT of homes for you. They GAVE them away. She banned the bit you could SEE, and kept the engine running. 21 brand new houses in a quiet Shropshire village. Worth £250,000 each. Built for local families. Handed to the asylum system instead, for up to 83 migrants, rent free. Rural. Back gardens. Next to where local children play. On a street where a British family had already bought and moved in. Residents weren't even told who was coming. They had to FOI their own council for the basics. How many. Who. Whether they're even families. Made to force it out in writing. Asked on live radio why our own homeless get nothing while this happens, Yvette Cooper didn’t deny it. She admitted it, said they should be in more appropriate accommodation. So she knows it’s wrong. Then she swerved. Reduce the system. Family reunion. Student visas. The system we inherited. We're saving a billion. Everything except the answer. So here's the answer she wouldn't give. Who decides it. Not your council. Not your MP. Not one person in that village. The Home Office decides, and tells a private firm, Serco, to place people. We know because Serco said it. The Home Office decides where people go, and instructs us accordingly. The people you vote for have no say. The people with the say, you never voted for. Why they pack them in. They're paid by the head. Twelve hundred pounds a year per person, a hundred a month per extra bed. Six in a house beats a family of four. That's how you get 83 people into 21 homes. Not compassion. Arithmetic. Who even chooses who comes. Under the new scheme, charities, universities and employers pick. Not you. Not your MP. Private groups hold the keys. Who gets rich. The man who owns one of these firms went from a caravan park to a billionaire. Porsches. Mayfair. Monaco. All from housing migrants on our money. His firm paid out twenty eight million in dividends in one year, almost all to him. One man, drawing more from our foreign aid budget than the whole of Ghana. He made so much his firm agreed to hand thirty two million back for breaching the profit cap. Money the Home Office still hasn't confirmed it's even collected. Here’s why they'll NEVER really stop it. The money we spend housing migrants here is counted as foreign aid. On paper, Britain feeding the world's poor. Really, filling a landlord's pockets in Monaco. Because it counts as aid, the more come, the better their books look. Count the ways they win. The contractor gets rich. The Treasury books four point three billion as aid that never leaves the country. Real aid to Africa is cut, the money stays here in Clearsprings and the consultancies, and you pay twice, in tax and in a home your kids will never get. The boats aren't a crisis to them. They're the business model. Forty thousand to leave, called a saving. Ten thousand to stay, called fairness, though their own figures say most will never earn enough to pay a penny. Now, ban the new-builds, and call it fixed. A different headline every month. The same engine underneath. Here’s the smoke, right on cue. They've said Stoke Heath won't go ahead. Sounds like a win. It isn't. They didn't stop it because it was wrong. They stopped it because it was SEEN. It went viral. It got too hot. So they killed the one estate the camera found, and kept every one it didn't. Its not just Shropshire. Over in Suffolk, migrants have already moved into four brand new townhouses worth £300,000, with en-suites, underfloor heating and electric car points, handed over rent free Flat blocks taken in Huddersfield, Chelmsford, Bournemouth. Same Home Office. Same contractor. Town after town, the ones no camera found. They won't even confirm the family already in Stoke Heath will be moved out. The ban only touches new-builds. The flat blocks stay. The HMOs stay. The contracts stay. Who really runs it. The Home Office behind all this was headed by Antonia Romeo, now Cabinet Secretary, the most powerful official in the country. She began her own career at a global consultancy, Oliver Wyman, a firm with a documented history of work in the refugee sector. Nobody voted for her. She was there before this government, and she'll be there after it. Ministers are the face. She's the engine. Three firms. Fifteen billion pounds of contracts. Break clauses that could end it tomorrow. Not one pulled. Yvette Cooper ran this and dodged it on air. Shabana Mahmood bans the one part that got caught and keeps the rest. Same engine. Two faces. The next face is already here. Whoever walks into Number 10 next can't break it either. Romeo stays. The contracts stay. They inherit it intact. Our own keep waiting. 1.3 million on the longest social housing queue in over a decade. The family that saved ten years. The homeless man born here. Passed over, in their own country, so the engine keeps running. You can respray a car. New paint, new bodywork, new badge. It still drives, because you never touched the engine. Same here. Change the government, the face, the party. The engine never moves. The Home Office. The contractor. The billionaire. The consultancy. The money. The one they never let you vote out. That's the engine and every few months they hand you a new face to be angry at, so you keep swinging at the paint. They didn't ban the machine. They banned the photo of it. Then stood at a podium and told us they'd fixed it. That's not a mistake. That's a decision. Stop aiming at the paint. Go for the engine.

BanksyCat

27,938 просмотров • 15 дней назад

I finally finished my Rust version of Mario Zechner's (Mario Zechner) excellent Pi Agent, which I made with his blessing and which is called pi_agent_rust. You can get it here: If you're not familiar with Pi, it's a minimalist and extensible agent harness (similar to Claude Code and Codex) and, among other uses, serves as the core agent harness inside the OpenClaw project. I say my Rust "version" instead of "port" because it's really quite different in how it's implemented for it to be called a port. Arguably, the incremental functionality in the implementation was more complex than the rest of the project combined. Still, it provides the same features and functionality as the original, and is proven to be compatible with hundreds of popular extensions to Pi (the conformance harness shows 224 out of 224 extensions working perfectly). But the way it's architected has some major changes. Pi Agent relies on node or bun to provide access to the filesystem and for various other tasks, and that is also how Pi's extension system works. I decided early on that I didn't want to do things that way. Instead, I wanted to integrate that functionality directly into the binary itself; that is, to provide equivalent functionality for everything that would normally be provided by node/bun in the original. I did this for several reasons: one, it's a lot more performant in terms of footprint and latency. On realistic end-to-end large-session workloads (not toy microbenchmarks), pi_agent_rust is now: - 4.95x faster than legacy Node and 2.80x faster than legacy Bun at 1mm-token session scale - 4.32x faster than legacy Node and 2.14x faster than legacy Bun at 5mm-token session scale - ~8x to ~13x lower RSS memory footprint in those same scenarios But the other reason is security and control: by handling everything internally in an end-to-end way, we can do all sorts of clever things to harden the system against insecure or malicious extensions. Those extensions no longer have direct access to the ambient filesystem: they now need to go through pi_agent_rust, and we can analyze extensions carefully before ever running them and also block things that look suspicious at runtime. In practice that means explicit capability-gated hostcalls, with policy/risk/quota enforcement and runtime telemetry/auditability. In order to do all this, I had to effectively build the missing runtime substrate from scratch in Rust, not just translate TypeScript syntax: - define and implement a typed hostcall ABI for extension->host interactions - build native Rust connectors for tool/exec/http/session/ui/events instead of ambient Node/Bun access - implement a compatibility/shim layer so real-world Pi extensions still behave correctly - add capability policy evaluation, runtime risk scoring, per-extension quotas, and audit telemetry on the execution path - wire the whole thing through structured concurrency (asupersync) so cancellation/lifetimes are deterministic and failure handling is explicit - build a conformance + benchmark harness large enough to validate behavior/perf across hundreds of extensions and realistic long-session workloads This was a full re-architecture of the execution model while preserving the Pi workflow and extension ecosystem. And indeed, this aspect of it dwarfs the entire rest of the project in size and complexity. To put hard numbers on that: the extension/runtime/security subsystem alone is now about 86.5k lines of Rust across src/extensions.rs (~48.1k), src/extensions_js.rs (~23.4k), src/extension_dispatcher.rs (~13.4k), and src/extension_index.rs (~1.7k), with roughly 2.5k callable units in just those files. For context, the original Pi coding-agent production code is about 27.4k lines total. So this one subsystem by itself is roughly 3.2x the size of the original harness, which is why calling this a “port” would seriously undersell what had to be built. And on top of that, pi_agent_rust introduces a bunch of genuinely new capabilities beyond the legacy harness, not just a faster core: - Security and enforcement are materially stronger at runtime: capability-gated hostcalls with explicit policy profiles (safe/balanced/permissive), per-extension trust lifecycle (pending -> acknowledged -> trusted -> killed), explicit kill-switch operations, and audited state transitions. - Shell execution mediation is deterministic and argument-aware: rule/feature-based risk scoring plus heredoc AST inspection (dcg_rule_hit, dcg_heredoc_hit) before spawn, instead of relying on coarse deny patterns. - Containment and forensics are first-class: tamper-evident runtime risk ledger tooling (verify/replay/calibrate), unified incident evidence bundles, and forced-compat controls that let you contain issues without disabling the whole extension system. - The extension runtime architecture is native: JS extensions run in embedded QuickJS with typed hostcall boundaries and Rust-native connectors for tool/exec/http/session/ui/events, plus compatibility shims for real-world legacy extensions. - Runtime behavior under load is explicitly engineered: deterministic hostcall reactor mesh, fast-lane vs compat-lane routing, and warm-isolate prewarm handoff for more predictable throughput and latency under contention. - Long-session reliability is upgraded: JSONL v3 sessions with indexed sidecar acceleration and optional SQLite-backed sessions, plus operational controls via --session-durability, --no-migrations, and migrate. - Provider and auth coverage are broader and more operationally explicit: native Anthropic/OpenAI (Chat + Responses)/Gemini/Cohere/Azure/Bedrock/Vertex/Copilot/GitLab plus large OpenAI-compatible routing; pi --list-providers currently shows 90 providers with aliases and required auth env keys. - Auth is not just API keys: OAuth (Anthropic/OpenAI Codex/Gemini CLI/Antigravity/Kimi/Copilot/GitLab plus extension-defined OAuth), AWS credential chains (Bedrock), service-key exchange (SAP AI Core), and bearer-token flows. - Operator tooling is stronger: pi doctor supports scoped checks (config, dirs, auth, shell, sessions, extensions), machine-readable output (--format json|markdown), and safe auto-remediation (--fix). - Extension/package lifecycle workflows are built in: install, remove, update, update-index, search, info, and list. I want to thank Mario for making a great harness and for not telling me to get lost when I asked him if he was OK with me porting it to Rust. I may give him a hard time in jest about not going "full clanker," but that doesn't mean that I don't respect his work a huge amount. PS: There could still be bugs. If you find some, please let me know in GitHub Issues and I'll fix them same day. There's always a tradeoff between perfect and getting stuff out the door and I felt like it was time to release this.
0:38

I finally finished my Rust version of Mario Zechner's (Mario Zechner) excellent Pi Agent, which I made with his blessing and which is called pi_agent_rust. You can get it here: If you're not familiar with Pi, it's a minimalist and extensible agent harness (similar to Claude Code and Codex) and, among other uses, serves as the core agent harness inside the OpenClaw project. I say my Rust "version" instead of "port" because it's really quite different in how it's implemented for it to be called a port. Arguably, the incremental functionality in the implementation was more complex than the rest of the project combined. Still, it provides the same features and functionality as the original, and is proven to be compatible with hundreds of popular extensions to Pi (the conformance harness shows 224 out of 224 extensions working perfectly). But the way it's architected has some major changes. Pi Agent relies on node or bun to provide access to the filesystem and for various other tasks, and that is also how Pi's extension system works. I decided early on that I didn't want to do things that way. Instead, I wanted to integrate that functionality directly into the binary itself; that is, to provide equivalent functionality for everything that would normally be provided by node/bun in the original. I did this for several reasons: one, it's a lot more performant in terms of footprint and latency. On realistic end-to-end large-session workloads (not toy microbenchmarks), pi_agent_rust is now: - 4.95x faster than legacy Node and 2.80x faster than legacy Bun at 1mm-token session scale - 4.32x faster than legacy Node and 2.14x faster than legacy Bun at 5mm-token session scale - ~8x to ~13x lower RSS memory footprint in those same scenarios But the other reason is security and control: by handling everything internally in an end-to-end way, we can do all sorts of clever things to harden the system against insecure or malicious extensions. Those extensions no longer have direct access to the ambient filesystem: they now need to go through pi_agent_rust, and we can analyze extensions carefully before ever running them and also block things that look suspicious at runtime. In practice that means explicit capability-gated hostcalls, with policy/risk/quota enforcement and runtime telemetry/auditability. In order to do all this, I had to effectively build the missing runtime substrate from scratch in Rust, not just translate TypeScript syntax: - define and implement a typed hostcall ABI for extension->host interactions - build native Rust connectors for tool/exec/http/session/ui/events instead of ambient Node/Bun access - implement a compatibility/shim layer so real-world Pi extensions still behave correctly - add capability policy evaluation, runtime risk scoring, per-extension quotas, and audit telemetry on the execution path - wire the whole thing through structured concurrency (asupersync) so cancellation/lifetimes are deterministic and failure handling is explicit - build a conformance + benchmark harness large enough to validate behavior/perf across hundreds of extensions and realistic long-session workloads This was a full re-architecture of the execution model while preserving the Pi workflow and extension ecosystem. And indeed, this aspect of it dwarfs the entire rest of the project in size and complexity. To put hard numbers on that: the extension/runtime/security subsystem alone is now about 86.5k lines of Rust across src/extensions.rs (~48.1k), src/extensions_js.rs (~23.4k), src/extension_dispatcher.rs (~13.4k), and src/extension_index.rs (~1.7k), with roughly 2.5k callable units in just those files. For context, the original Pi coding-agent production code is about 27.4k lines total. So this one subsystem by itself is roughly 3.2x the size of the original harness, which is why calling this a “port” would seriously undersell what had to be built. And on top of that, pi_agent_rust introduces a bunch of genuinely new capabilities beyond the legacy harness, not just a faster core: - Security and enforcement are materially stronger at runtime: capability-gated hostcalls with explicit policy profiles (safe/balanced/permissive), per-extension trust lifecycle (pending -> acknowledged -> trusted -> killed), explicit kill-switch operations, and audited state transitions. - Shell execution mediation is deterministic and argument-aware: rule/feature-based risk scoring plus heredoc AST inspection (dcg_rule_hit, dcg_heredoc_hit) before spawn, instead of relying on coarse deny patterns. - Containment and forensics are first-class: tamper-evident runtime risk ledger tooling (verify/replay/calibrate), unified incident evidence bundles, and forced-compat controls that let you contain issues without disabling the whole extension system. - The extension runtime architecture is native: JS extensions run in embedded QuickJS with typed hostcall boundaries and Rust-native connectors for tool/exec/http/session/ui/events, plus compatibility shims for real-world legacy extensions. - Runtime behavior under load is explicitly engineered: deterministic hostcall reactor mesh, fast-lane vs compat-lane routing, and warm-isolate prewarm handoff for more predictable throughput and latency under contention. - Long-session reliability is upgraded: JSONL v3 sessions with indexed sidecar acceleration and optional SQLite-backed sessions, plus operational controls via --session-durability, --no-migrations, and migrate. - Provider and auth coverage are broader and more operationally explicit: native Anthropic/OpenAI (Chat + Responses)/Gemini/Cohere/Azure/Bedrock/Vertex/Copilot/GitLab plus large OpenAI-compatible routing; pi --list-providers currently shows 90 providers with aliases and required auth env keys. - Auth is not just API keys: OAuth (Anthropic/OpenAI Codex/Gemini CLI/Antigravity/Kimi/Copilot/GitLab plus extension-defined OAuth), AWS credential chains (Bedrock), service-key exchange (SAP AI Core), and bearer-token flows. - Operator tooling is stronger: pi doctor supports scoped checks (config, dirs, auth, shell, sessions, extensions), machine-readable output (--format json|markdown), and safe auto-remediation (--fix). - Extension/package lifecycle workflows are built in: install, remove, update, update-index, search, info, and list. I want to thank Mario for making a great harness and for not telling me to get lost when I asked him if he was OK with me porting it to Rust. I may give him a hard time in jest about not going "full clanker," but that doesn't mean that I don't respect his work a huge amount. PS: There could still be bugs. If you find some, please let me know in GitHub Issues and I'll fix them same day. There's always a tradeoff between perfect and getting stuff out the door and I felt like it was time to release this.

Jeffrey Emanuel

116,929 просмотров • 4 месяцев назад

RFK Jr.’s ‘moonshot program’ to end the overdose crisis with recovery centers in every rural and urban community: “No screens, you’ll grow your own food, and you’ll come out of it will a skill.” “Large numbers of [kids] are on mood-control drugs, on Adderall very early on to control their behavior in schools, and that starts a cycle of addiction. A large percentage of them are on SSRI drugs, on benzos, which actually amplify suicidal effects. There’s black box warnings on those products that say ‘suicidal and homicidal ideation’ as one of the side effects. We have 106,000 kids who died of overdoses last year. Twice the number that died during the twenty-year Vietnam War. We have two Vietnams every year killing our kids. My ‘moonshot’ program is to make recovery available for free to every kid in our country... I’m going to put in all rural areas and in urban areas rehabilitation farms. Places where people can go to get spiritual renewal, to get off of drugs, not only illegal drugs but also SSRIs and these psychiatric drugs, and learn how to be part of a community. There will be no screens there, no computers, no cell phones, because that’s part of the addiction and the issue. They’ll grow their own food, organic food, and you will come out of that with a skill. The model for this is something that I saw, that a family member of mine had a transformative experience with, called San Patrignano in Italy. There’s 2,000 kids, the only commitment you have to make is that you’re going to stay there for four years. It is absolutely free, they grow their own food, they have a 500-acre vineyard where they make some of the finest wines in Europe, they have a bakery, they sell baked goods all over Italy, they have an apparel shop, factories, they have wallpaper factories… Every kid comes out of there with a skill. I visited three weeks ago a project that is modeled on San Patrignano in Salt Lake City, Utah. It’s five urban blocks where there’s 300 convicted felons who’ve come out of prison, many of them drug addicts, they’re given the same kind of regimen. This place, which is called The Other Side, runs the biggest storage center in Utah. It runs the biggest moving company… it has a used clothing shop that’s bigger than two Walmarts. It’s hugely successful, and it supports itself through these economic enterprises. When you go in there, you see a list on the wall of hundreds of businesses that are waiting for graduates of this program… almost all of the businesses in Utah are desperate to get people who graduate this program because they’re known to be the most responsible workforce. That is a model that we can now roll out across the country, and I’m going to fund this by moving marijuana off of Schedule I. Cannabis is legal in many states, but the people who are selling it… cannot pay federal taxes and they can’t legally put that money in a bank.”
6:37

RFK Jr.’s ‘moonshot program’ to end the overdose crisis with recovery centers in every rural and urban community: “No screens, you’ll grow your own food, and you’ll come out of it will a skill.” “Large numbers of [kids] are on mood-control drugs, on Adderall very early on to control their behavior in schools, and that starts a cycle of addiction. A large percentage of them are on SSRI drugs, on benzos, which actually amplify suicidal effects. There’s black box warnings on those products that say ‘suicidal and homicidal ideation’ as one of the side effects. We have 106,000 kids who died of overdoses last year. Twice the number that died during the twenty-year Vietnam War. We have two Vietnams every year killing our kids. My ‘moonshot’ program is to make recovery available for free to every kid in our country... I’m going to put in all rural areas and in urban areas rehabilitation farms. Places where people can go to get spiritual renewal, to get off of drugs, not only illegal drugs but also SSRIs and these psychiatric drugs, and learn how to be part of a community. There will be no screens there, no computers, no cell phones, because that’s part of the addiction and the issue. They’ll grow their own food, organic food, and you will come out of that with a skill. The model for this is something that I saw, that a family member of mine had a transformative experience with, called San Patrignano in Italy. There’s 2,000 kids, the only commitment you have to make is that you’re going to stay there for four years. It is absolutely free, they grow their own food, they have a 500-acre vineyard where they make some of the finest wines in Europe, they have a bakery, they sell baked goods all over Italy, they have an apparel shop, factories, they have wallpaper factories… Every kid comes out of there with a skill. I visited three weeks ago a project that is modeled on San Patrignano in Salt Lake City, Utah. It’s five urban blocks where there’s 300 convicted felons who’ve come out of prison, many of them drug addicts, they’re given the same kind of regimen. This place, which is called The Other Side, runs the biggest storage center in Utah. It runs the biggest moving company… it has a used clothing shop that’s bigger than two Walmarts. It’s hugely successful, and it supports itself through these economic enterprises. When you go in there, you see a list on the wall of hundreds of businesses that are waiting for graduates of this program… almost all of the businesses in Utah are desperate to get people who graduate this program because they’re known to be the most responsible workforce. That is a model that we can now roll out across the country, and I’m going to fund this by moving marijuana off of Schedule I. Cannabis is legal in many states, but the people who are selling it… cannot pay federal taxes and they can’t legally put that money in a bank.”

Holden Culotta

172,238 просмотров • 2 лет назад