Загрузка видео...

Не удалось загрузить видео

На главную

Same text. Two privacy filters. OpenAI's model catches 8 categories. OpenMed catches 55+: medical record numbers, blood type, API keys, financial codes, demographics. Trained on Nemotron data by Nvidia. All on-device. All open-source. Coming soon! What's missing?

MaziyarPanahi's profile picture

Maziyar PANAHI

19,205 subscribers

122,309 просмотров • 5 месяцев назад •via X (Twitter)

Здоровье и велнесНаука и технологии
LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

THIS IS F**KING DANGEROUS THREE OPEN SOURCE AI CYBERSECURITY STACK MOST PEOPLE DONT KNOW EXISTS METATRON: point it at a target, it recon's, analyzes, and reports on its own —> fully OFFLINE , zero API keys. NUCLEI : describe a vulnerability in plain English, it writes and runs the actual detection template for you. GARAK BY (NVIDIA): throws thousands of known prompt injection and jailbreak attempts at an LLM until one breaks through. Recon —> detection —> and AI red teaming three layers of the same problem, all open source. Link to access the REPO’S below
0:19

THIS IS F**KING DANGEROUS THREE OPEN SOURCE AI CYBERSECURITY STACK MOST PEOPLE DONT KNOW EXISTS METATRON: point it at a target, it recon's, analyzes, and reports on its own —> fully OFFLINE , zero API keys. NUCLEI : describe a vulnerability in plain English, it writes and runs the actual detection template for you. GARAK BY (NVIDIA): throws thousands of known prompt injection and jailbreak attempts at an LLM until one breaks through. Recon —> detection —> and AI red teaming three layers of the same problem, all open source. Link to access the REPO’S below

MAX

105,425 просмотров • 1 месяц назад

Goldman pays $27,000 per seat for a Bloomberg Terminal. I found 10 open source tools on GitHub that replicate almost all of it for free. Retail investors have never had this much firepower. Bookmark & Repost this one: 1. OpenBB Stocks, options, crypto, forex, and macro data in one research platform. Build your own dashboards, reports, and AI analysts on top of it. The OG of open source finance. 50K+ stars. 2. FinceptTerminal A full financial terminal: global market data, advanced charts, economic indicators, portfolio analysis, and AI research tools. Windows, Mac, and Linux. 3. Neuberg 516 drag-and-drop panels covering equities, bonds, commodities, currencies, credit, and macro. Even connects to Alpaca, Hyperliquid, and Polymarket so you can trade from the terminal itself. 4. Qlib (by Microsoft) An open source AI platform for quant investing. Train ML models, discover signals, backtest strategies, and build portfolios with the same workflow a quant desk uses. 5. FinRobot An AI equity research team on your laptop. Its agents read financial statements, build DCF valuations, debate bull vs bear cases, and generate full investment reports. 6. EdgarTools Turns the SEC database into something humans can actually use. Pull 10-Ks, 10-Qs, insider trades, executive pay, and hedge fund holdings going back to 1994. 7. LEAN (by QuantConnect) An institutional-grade engine for trading algorithms. Write strategies in Python or C#, backtest on decades of data, then connect to real brokers and go live. 8. FinanceToolkit 200+ financial ratios, valuation models, risk metrics, and economic indicators. Works on stocks, ETFs, options, currencies, commodities, and crypto from Python. 9. Ghostfolio A private wealth dashboard for stocks, ETFs, and crypto across all your accounts. Performance, allocation, diversification. Your data never leaves your machine. 10. OpenTerminalUI A self-hosted trading terminal: pro charts, screeners, options chains with live Greeks, portfolio optimization, backtesting, and an AI research agent. Runs entirely on your own hardware. Bloomberg spent 40 years building a $27,000/year moat. Open source is draining it one repo at a time. The software is free. Some live data feeds need your own API keys, but the barrier is now effort, not money. If you want the exact workflows we use to stack these tools with AI, join the AIBullss Discord:
0:13

Goldman pays $27,000 per seat for a Bloomberg Terminal. I found 10 open source tools on GitHub that replicate almost all of it for free. Retail investors have never had this much firepower. Bookmark & Repost this one: 1. OpenBB Stocks, options, crypto, forex, and macro data in one research platform. Build your own dashboards, reports, and AI analysts on top of it. The OG of open source finance. 50K+ stars. 2. FinceptTerminal A full financial terminal: global market data, advanced charts, economic indicators, portfolio analysis, and AI research tools. Windows, Mac, and Linux. 3. Neuberg 516 drag-and-drop panels covering equities, bonds, commodities, currencies, credit, and macro. Even connects to Alpaca, Hyperliquid, and Polymarket so you can trade from the terminal itself. 4. Qlib (by Microsoft) An open source AI platform for quant investing. Train ML models, discover signals, backtest strategies, and build portfolios with the same workflow a quant desk uses. 5. FinRobot An AI equity research team on your laptop. Its agents read financial statements, build DCF valuations, debate bull vs bear cases, and generate full investment reports. 6. EdgarTools Turns the SEC database into something humans can actually use. Pull 10-Ks, 10-Qs, insider trades, executive pay, and hedge fund holdings going back to 1994. 7. LEAN (by QuantConnect) An institutional-grade engine for trading algorithms. Write strategies in Python or C#, backtest on decades of data, then connect to real brokers and go live. 8. FinanceToolkit 200+ financial ratios, valuation models, risk metrics, and economic indicators. Works on stocks, ETFs, options, currencies, commodities, and crypto from Python. 9. Ghostfolio A private wealth dashboard for stocks, ETFs, and crypto across all your accounts. Performance, allocation, diversification. Your data never leaves your machine. 10. OpenTerminalUI A self-hosted trading terminal: pro charts, screeners, options chains with live Greeks, portfolio optimization, backtesting, and an AI research agent. Runs entirely on your own hardware. Bloomberg spent 40 years building a $27,000/year moat. Open source is draining it one repo at a time. The software is free. Some live data feeds need your own API keys, but the barrier is now effort, not money. If you want the exact workflows we use to stack these tools with AI, join the AIBullss Discord:

AI Bulls

20,605 просмотров • 1 месяц назад

encrypt every chat with this open sourced repo 18 year old just killed the encryption problem it's called conversation-steganography you type a secret and a local AI buries it inside a dead boring chat about coffee, etc. then you send that through any normal messenger your friend runs the same tool and the small talk unfolds back into the real words > the secret is locked with AES-SIV before it ever becomes text > the cover message is written by a model on your own machine so nothing touches the cloud > it rides through whatsapp telegram imessage even plain email > every message is cryptographically chained so tampering shows instantly whatsapp sees you planning dinner your isp logs small talk the real sentence was never in the data at all quietly one of the most dangerous repos i have seen all month repo in reply
0:16

encrypt every chat with this open sourced repo 18 year old just killed the encryption problem it's called conversation-steganography you type a secret and a local AI buries it inside a dead boring chat about coffee, etc. then you send that through any normal messenger your friend runs the same tool and the small talk unfolds back into the real words > the secret is locked with AES-SIV before it ever becomes text > the cover message is written by a model on your own machine so nothing touches the cloud > it rides through whatsapp telegram imessage even plain email > every message is cryptographically chained so tampering shows instantly whatsapp sees you planning dinner your isp logs small talk the real sentence was never in the data at all quietly one of the most dangerous repos i have seen all month repo in reply

savip.

83,651 просмотров • 2 месяцев назад

Your LG OLED is watching your living room while you sleep. Gamers Nexus just ran 135 minutes of bench tests. Retail LG OLED TVs. Including the G5. Microphone stays active in standby. Screen is dark. Remote is down. Microphone is still on. Recording clear audio. Storing it locally. Uploading it when the internet comes back. The TV was disconnected from Ethernet. Still kept recording. The researchers put audio near the mic. The TV logged it. Stored the file. Uploaded it the moment connectivity returned. Plain text voice logs sitting on the device. LG says it has sold 216 million smart TVs globally. LG's advertising arm claims access to 363 million secondary devices in the US. Secondary. Meaning devices that aren't the TV. Your phone. Your partner's smartwatch. Your kid's laptop. The TV scans your entire network in standby. Maps every device. Gets their IP addresses. Logs nearby Wi-Fi networks. Signal strengths. Location data. Then feeds it to LG Ad Solutions. Automatic Content Recognition catches what's on screen. Even your HDMI input. Your Apple TV. Your gaming console. All logged. But the microphone in standby is the part that hits different. Because standby means off to you. To LG it means the listening device is still powered. Just not telling you. Researchers found remote code execution vulnerabilities in webOS. So your network-scanning, audio-recording, data-harvesting television is also potentially hackable by anyone on your Wi-Fi. LG says nothing. No comment. No response. No explanation. Just microphones in 216 million living rooms. All quietly working while the owner thinks the TV is asleep.
0:13

Your LG OLED is watching your living room while you sleep. Gamers Nexus just ran 135 minutes of bench tests. Retail LG OLED TVs. Including the G5. Microphone stays active in standby. Screen is dark. Remote is down. Microphone is still on. Recording clear audio. Storing it locally. Uploading it when the internet comes back. The TV was disconnected from Ethernet. Still kept recording. The researchers put audio near the mic. The TV logged it. Stored the file. Uploaded it the moment connectivity returned. Plain text voice logs sitting on the device. LG says it has sold 216 million smart TVs globally. LG's advertising arm claims access to 363 million secondary devices in the US. Secondary. Meaning devices that aren't the TV. Your phone. Your partner's smartwatch. Your kid's laptop. The TV scans your entire network in standby. Maps every device. Gets their IP addresses. Logs nearby Wi-Fi networks. Signal strengths. Location data. Then feeds it to LG Ad Solutions. Automatic Content Recognition catches what's on screen. Even your HDMI input. Your Apple TV. Your gaming console. All logged. But the microphone in standby is the part that hits different. Because standby means off to you. To LG it means the listening device is still powered. Just not telling you. Researchers found remote code execution vulnerabilities in webOS. So your network-scanning, audio-recording, data-harvesting television is also potentially hackable by anyone on your Wi-Fi. LG says nothing. No comment. No response. No explanation. Just microphones in 216 million living rooms. All quietly working while the owner thinks the TV is asleep.

Abdulkadir | Cybersecurity

2,200,406 просмотров • 18 дней назад

DarkFi superapp v0.3-alpha is here, “anon”. DarkIRC chat is live. Payments, DAO and trading coming as addons. One native application to rule them all. This is the crypto infrastructure you were promised. These are the user-facing apps OGs dreamed of. Why the superapp matters: Anonymous digital spaces are the pre-condition for real-world autonomy in 2025. When developers can build without exposure, when users can coordinate without surveillance, when governance happens in zero-knowledge - then you create a space for experimentation and growth. Anonymous DAOs make decisions without revealing voter identities or treasury contents. Anonymous communication enables collaboration across jurisdictions without compromising participants. This is the infrastructure for sovereignty. DarkFi built one application that runs identically on Linux, Android, Mac and Windows. Developers write once, deploy everywhere with zero exposure. Users get the same experience on every device. Third-party builders get full API access without asking permission. Software runs locally. What's in v0.3: - Android text input fixed. - Graphics rendering corrected to eliminate glitching. - Battery consumption reduced on mobile. - UI improvements with animated background. - Second stabilization release since 0.2-alpha (v0.4 coming soon with additional features). Let there be Dark!
0:12

DarkFi superapp v0.3-alpha is here, “anon”. DarkIRC chat is live. Payments, DAO and trading coming as addons. One native application to rule them all. This is the crypto infrastructure you were promised. These are the user-facing apps OGs dreamed of. Why the superapp matters: Anonymous digital spaces are the pre-condition for real-world autonomy in 2025. When developers can build without exposure, when users can coordinate without surveillance, when governance happens in zero-knowledge - then you create a space for experimentation and growth. Anonymous DAOs make decisions without revealing voter identities or treasury contents. Anonymous communication enables collaboration across jurisdictions without compromising participants. This is the infrastructure for sovereignty. DarkFi built one application that runs identically on Linux, Android, Mac and Windows. Developers write once, deploy everywhere with zero exposure. Users get the same experience on every device. Third-party builders get full API access without asking permission. Software runs locally. What's in v0.3: - Android text input fixed. - Graphics rendering corrected to eliminate glitching. - Battery consumption reduced on mobile. - UI improvements with animated background. - Second stabilization release since 0.2-alpha (v0.4 coming soon with additional features). Let there be Dark!

DarkFi Squad

19,827 просмотров • 10 месяцев назад

you're paying $20/mo for something your $500 GPU can already do. Gemma 4 26B A4B QAT MoE + Hermes Agent running on a single RTX 4060 (8GB VRAM). Built a vision capable, 100% free, 100% local, private AI assistant that lives in my Chrome browser. No API keys. No cloud. No subscriptions. 100% vibe coded. 0% handholding. It has full context of whatever's on my screen can answer questions, summarize pages, extract data, and see images. Same local model handles everything, no external calls, ever. keep reading for the model and hermes agent tips i learnt while building this locally. Here's the exact setup for anyone running local LLMs on 6-8 GB VRAM: llama.cpp server flags (on my NVIDIA RTX 4060 8gb VRAM): -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --cache-type-k q8_0 --cache-type-v q8_0 -c 150000 --port 8080 Throughput with quantization: Prefill: 200-250 tokens/sec Decode: 20-25 tokens/sec reduce context if oom on 6 gb vram card. Key learnings: - Quantize KV cache to q8 for faster prefill/decode. Prefill goes from 100-150 (unquantized) to 200-250 tok/s (q8). - But watch out, once actual context grows past ~50k tokens on high entropy workloads, q8 KV quantization can cause hallucinations. Low entropy workloads are mostly unaffected. If you see it happening, drop the quantization. This is common across all local models. - In Hermes Agent settings -> Memory & Context, bump compression threshold from default 0.5 to 0.7. Default triggers way too frequent context compression and eats time. Up next: add persistent memory, web search, tool calling, streaming output and whatever you suggest. Running a 26B MoE with vision + 150k context window on 8GB VRAM would've sounded impossible 6 months ago. Works the same on the NVIDIA RTX 3060 Ti, 3070, 4060 Ti, 5060, 2080, or any 8GB card. VRAM is the only requirement. Local AI agents are closer than people think. You just need to know where the knobs are. Model's Unsloth quant hugging face link in the comments. Have you tried Hermes agent by Nous Research yet? What are you building with local LLMs? Drop it below, let's see what this community is shipping.
0:19

you're paying $20/mo for something your $500 GPU can already do. Gemma 4 26B A4B QAT MoE + Hermes Agent running on a single RTX 4060 (8GB VRAM). Built a vision capable, 100% free, 100% local, private AI assistant that lives in my Chrome browser. No API keys. No cloud. No subscriptions. 100% vibe coded. 0% handholding. It has full context of whatever's on my screen can answer questions, summarize pages, extract data, and see images. Same local model handles everything, no external calls, ever. keep reading for the model and hermes agent tips i learnt while building this locally. Here's the exact setup for anyone running local LLMs on 6-8 GB VRAM: llama.cpp server flags (on my NVIDIA RTX 4060 8gb VRAM): -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --cache-type-k q8_0 --cache-type-v q8_0 -c 150000 --port 8080 Throughput with quantization: Prefill: 200-250 tokens/sec Decode: 20-25 tokens/sec reduce context if oom on 6 gb vram card. Key learnings: - Quantize KV cache to q8 for faster prefill/decode. Prefill goes from 100-150 (unquantized) to 200-250 tok/s (q8). - But watch out, once actual context grows past ~50k tokens on high entropy workloads, q8 KV quantization can cause hallucinations. Low entropy workloads are mostly unaffected. If you see it happening, drop the quantization. This is common across all local models. - In Hermes Agent settings -> Memory & Context, bump compression threshold from default 0.5 to 0.7. Default triggers way too frequent context compression and eats time. Up next: add persistent memory, web search, tool calling, streaming output and whatever you suggest. Running a 26B MoE with vision + 150k context window on 8GB VRAM would've sounded impossible 6 months ago. Works the same on the NVIDIA RTX 3060 Ti, 3070, 4060 Ti, 5060, 2080, or any 8GB card. VRAM is the only requirement. Local AI agents are closer than people think. You just need to know where the knobs are. Model's Unsloth quant hugging face link in the comments. Have you tried Hermes agent by Nous Research yet? What are you building with local LLMs? Drop it below, let's see what this community is shipping.

Alok

36,691 просмотров • 2 месяцев назад

SOMEONE BUILT A WAY TO SEND FILES BETWEEN TWO PHONES WITH NO WIFI, NO BLUETOOTH AND NO NETWORK AT ALL it works with just a screen flashing qr codes at a camera one device plays an endless stream of animated qr codes, the other one points its camera at the screen and rebuilds the file out of the light no pairing, no app, no shared network, nothing but the camera > its pulling about 129 kb/s, a full 2mb image transfers in seconds > theres no back channel, the receiver can never ask for a missed frame, so it uses fountain codes, every frame is a scrambled mix of the file rather than one piece of it > that means the receiver just collects any frames it can catch in any order, and once it has enough it can reconstruct the whole file > dropping frames costs you a little time but never breaks the transfer > you can point the camera at a stream thats already running and it just locks on mid flight he was building a cached mp3 player because he was sick of ads and bad streaming quality. he wanted a way to send tracks phone to phone without both phones being on the same network, and realized flashing qr codes could carry the data he built the whole working prototype in one night with claude code its open source and already at 375 stars
0:15

SOMEONE BUILT A WAY TO SEND FILES BETWEEN TWO PHONES WITH NO WIFI, NO BLUETOOTH AND NO NETWORK AT ALL it works with just a screen flashing qr codes at a camera one device plays an endless stream of animated qr codes, the other one points its camera at the screen and rebuilds the file out of the light no pairing, no app, no shared network, nothing but the camera > its pulling about 129 kb/s, a full 2mb image transfers in seconds > theres no back channel, the receiver can never ask for a missed frame, so it uses fountain codes, every frame is a scrambled mix of the file rather than one piece of it > that means the receiver just collects any frames it can catch in any order, and once it has enough it can reconstruct the whole file > dropping frames costs you a little time but never breaks the transfer > you can point the camera at a stream thats already running and it just locks on mid flight he was building a cached mp3 player because he was sick of ads and bad streaming quality. he wanted a way to send tracks phone to phone without both phones being on the same network, and realized flashing qr codes could carry the data he built the whole working prototype in one night with claude code its open source and already at 375 stars

Om Patel

951,210 просмотров • 1 месяц назад

Introducing Pods Hyperspace Pods lets a small group of people - a family, a startup, a few friends, to pool their laptops and desktops into one AI cluster. Everyone installs the CLI, someone creates a pod, shares an invite link, and the machines form a mesh. Models like Qwen 3.5 32B or GLM-5 Turbo that need more memory than any single laptop has get automatically sharded across the group's devices - layers split proportionally, inference pipelined through the ring. From the outside it looks like one OpenAI-compatible API endpoint with a pk_* key that drops straight into your AI tools and products. No configuration beyond pasting the key and changing the base URL. A team of five paying for cloud AI burns $500–2,000 a month on API calls. The same team's existing machines can serve Qwen 3.5 (competitive on SWE-bench) and GLM-5 Turbo (#1 on BrowseComp for tool-calling and web research) for free - the hardware is already on their desks. When a query genuinely needs a frontier model nobody has locally, the pod falls back to cloud at wholesale rates from a shared treasury. But for the daily work - code reviews, refactors, research, drafting - local models handle it and nobody gets billed. And when it is idle, you can rent out your pod on the compute marketplace, with fine-grained permissions for access management. There's no central server involved in inference. Prompts go from your machine to your pod members' machines and back: all of this enabled by the fully peer-to-peer Hyperspace network. Pod state - who's a member, which API keys are valid, how much treasury is left - is replicated across members with consensus, so the whole thing works on a local network. Members behind home routers don't need port forwarding either. The practical setup for most pods is three models covering different jobs: Qwen 3.5 32B for code and reasoning, GLM-5 Turbo for browsing and research, Gemma 4 for fast lightweight tasks. All running on hardware you already own. Pods ship today in Hyperspace v5.19. Model sharding, API keys, treasury, and Raft coordinator are all live. What Makes This Different - No middleman. Your prompts travel from your IDE to your pod members' hardware and back. There is no server in between reading your data. - No vendor lock-in. Pod membership, API keys, and treasury are replicated across your own machines using Raft consensus. If the internet goes down, your local network keeps working. There is no database in someone else's cloud that your pod depends on. - Automatic sharding. You don't configure layer ranges or calculate VRAM budgets. Tell the pod which model you want. It figures out how to split it across whatever hardware is online. - Real NAT traversal. Your friend behind a home router with a dynamic IP? Works. No VPN, no Tailscale, no port forwarding. The nodes handle it. - Free when local. This is the part that matters most. Cloud AI bills scale with usage. Pod inference on local hardware scales with nothing. The marginal cost of your 10,000th prompt is the electricity your laptop was already using. Coming soon: - Pod federation: pods form alliances with other pods. - Marketplace: pods with spare capacity can sell inference to other pods.
0:19

Introducing Pods Hyperspace Pods lets a small group of people - a family, a startup, a few friends, to pool their laptops and desktops into one AI cluster. Everyone installs the CLI, someone creates a pod, shares an invite link, and the machines form a mesh. Models like Qwen 3.5 32B or GLM-5 Turbo that need more memory than any single laptop has get automatically sharded across the group's devices - layers split proportionally, inference pipelined through the ring. From the outside it looks like one OpenAI-compatible API endpoint with a pk_* key that drops straight into your AI tools and products. No configuration beyond pasting the key and changing the base URL. A team of five paying for cloud AI burns $500–2,000 a month on API calls. The same team's existing machines can serve Qwen 3.5 (competitive on SWE-bench) and GLM-5 Turbo (#1 on BrowseComp for tool-calling and web research) for free - the hardware is already on their desks. When a query genuinely needs a frontier model nobody has locally, the pod falls back to cloud at wholesale rates from a shared treasury. But for the daily work - code reviews, refactors, research, drafting - local models handle it and nobody gets billed. And when it is idle, you can rent out your pod on the compute marketplace, with fine-grained permissions for access management. There's no central server involved in inference. Prompts go from your machine to your pod members' machines and back: all of this enabled by the fully peer-to-peer Hyperspace network. Pod state - who's a member, which API keys are valid, how much treasury is left - is replicated across members with consensus, so the whole thing works on a local network. Members behind home routers don't need port forwarding either. The practical setup for most pods is three models covering different jobs: Qwen 3.5 32B for code and reasoning, GLM-5 Turbo for browsing and research, Gemma 4 for fast lightweight tasks. All running on hardware you already own. Pods ship today in Hyperspace v5.19. Model sharding, API keys, treasury, and Raft coordinator are all live. What Makes This Different - No middleman. Your prompts travel from your IDE to your pod members' hardware and back. There is no server in between reading your data. - No vendor lock-in. Pod membership, API keys, and treasury are replicated across your own machines using Raft consensus. If the internet goes down, your local network keeps working. There is no database in someone else's cloud that your pod depends on. - Automatic sharding. You don't configure layer ranges or calculate VRAM budgets. Tell the pod which model you want. It figures out how to split it across whatever hardware is online. - Real NAT traversal. Your friend behind a home router with a dynamic IP? Works. No VPN, no Tailscale, no port forwarding. The nodes handle it. - Free when local. This is the part that matters most. Cloud AI bills scale with usage. Pod inference on local hardware scales with nothing. The marginal cost of your 10,000th prompt is the electricity your laptop was already using. Coming soon: - Pod federation: pods form alliances with other pods. - Marketplace: pods with spare capacity can sell inference to other pods.

Varun

309,792 просмотров • 5 месяцев назад

AI token usage is up 10x in 7 months, compounding 40%/MONTH! There is NO BUBBLE when demand is STILL accelerating And this is just OpenRouter, it doesn't count the labs direct token usage and APIs But here's what's interesting about these numbers, the demand is coming from everywhere at once US models (OpenAI, Anthropic, Google) keep growing, while Chinese open weight models (DeepSeek, Tencent, Xiaomi, Minimax) grew even faster and now drive over 60% of usage on OpenRouter Closed source and open source both compounding at the same time. This is literally the best case scenario for AI Infra investors It means both frontier model tokens and cheaper tokens have product market fit. This means the application layer is finding ways to use both and generate ROI with both types Demand for tokens IS demand for compute. This is why SpaceX is looking to build 10GW of compute by next year, because the demand is clearly here Now combine this demand set up, with NVIDIA yesterday announcing financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize over $500 billion of third party capital for AI infrastructure And Jensen has said publicly he expects $3 to $4 TRILLION of AI infrastructure spend by 2030 The build out will have to continue for a lot longer than the market is expecting, that is very clear to me. Don't let this consolidation period in AI infra stocks shake you out, they will have their moment again and take their next leg higher p.s. if you want to see how im investing in this, you can track my real-time portfolio and the research of all 5 Milk Road PRO analysts with live trade notifications, and it's just $1 to try it out (insane price just to check it out). Learn more here: Good luck out there!
0:19

AI token usage is up 10x in 7 months, compounding 40%/MONTH! There is NO BUBBLE when demand is STILL accelerating And this is just OpenRouter, it doesn't count the labs direct token usage and APIs But here's what's interesting about these numbers, the demand is coming from everywhere at once US models (OpenAI, Anthropic, Google) keep growing, while Chinese open weight models (DeepSeek, Tencent, Xiaomi, Minimax) grew even faster and now drive over 60% of usage on OpenRouter Closed source and open source both compounding at the same time. This is literally the best case scenario for AI Infra investors It means both frontier model tokens and cheaper tokens have product market fit. This means the application layer is finding ways to use both and generate ROI with both types Demand for tokens IS demand for compute. This is why SpaceX is looking to build 10GW of compute by next year, because the demand is clearly here Now combine this demand set up, with NVIDIA yesterday announcing financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize over $500 billion of third party capital for AI infrastructure And Jensen has said publicly he expects $3 to $4 TRILLION of AI infrastructure spend by 2030 The build out will have to continue for a lot longer than the market is expecting, that is very clear to me. Don't let this consolidation period in AI infra stocks shake you out, they will have their moment again and take their next leg higher p.s. if you want to see how im investing in this, you can track my real-time portfolio and the research of all 5 Milk Road PRO analysts with live trade notifications, and it's just $1 to try it out (insane price just to check it out). Learn more here: Good luck out there!

Kyle Reidhead | Milk Road

28,320 просмотров • 1 месяц назад

llama.cpp isn't just for text LLMs anymore. Pure C++ zero shot voice cloning just officially landed in mainline. Text generation was only step one. If you’re building autonomous local AI agents, real time voice assistants, or edge workflows, instant low latency audio is the missing piece. Thanks to PR #26254, Alibaba’s state of the art Qwen3 TTS model family is now natively supported directly inside the llama.cpp repository under the multimodal (mtmd) framework. No Python bloat. No massive PyTorch CUDA overhead. Just raw, hyper optimized C++ running GGUF voice weights. Here is why this native update is a massive deal for the open source local AI stack: # Multimodal Architecture (.gguf + mmproj) Qwen3-TTS splits the workload between the base language model backbone and a multimodal projection adapter. llama.cpp handles this using the llama-tts binary, mapping the text model alongside its --mmproj projector to process audio tokens seamlessly. # Zero Shot Voice Cloning in Seconds You don't need fine tuning or massive dataset training. Feed the C++ engine a single 5 to 10 second .wav audio sample using the --tts-speaker-file flag, and it accurately clones the exact timbre, tone, and accent on the fly. # Real World T4 GPU Benchmark & Resource FootprintRunning the 1.7B Base model in 8-bit quantization (Q8_0): - VRAM Footprint: ~7 GB peak VRAM during active zero-shot cloning. - Audio Quality: Studio grade, natural-sounding voice output in seconds. • - Execution: Direct execution via native compiled binaries or sub process calls. # Coming Next to llama-server (PR #26603) Beyond CLI execution, a native POST /tts HTTP endpoint is currently being added to llama-server, which will soon allow you to trigger voice generation directly via standard REST API requests! # quick note on Colab compilation: Because this code was merged into mainline very recently, pre-built third-party binaries haven't fully caught up yet. Compiling llama-tts directly from source on Google Colab's free CPU instance can take about 1 hour (or ~1-2 minutes if targeting single GPU arch like -DCMAKE_CUDA_ARCHITECTURES=75). Be patient during the build step, or compile it locally on your own rig for instant execution! To test this out yourself, I built a zero config Google Colab notebook that compiles llama.cpp, downloads the Q8_0 GGUF files from HuggingFace, and spins up an interactive Gradio Studio UI so you can record/upload 3 second clips and clone voices in real time. Stop sleeping on native C++ audio. The era of bulky Python audio pipelines is officially over. Links to the free Google Colab notebook and the official ggml org GGUF HuggingFace model repository are in the replies below! available in q4 and q8 both variants, 1 GB and 1.85 GBs respectively (requires additional ~500MB mmproj gguf) Are you building local voice agents yet? What does your current audio stack look like? Drop your setups below!
0:12

llama.cpp isn't just for text LLMs anymore. Pure C++ zero shot voice cloning just officially landed in mainline. Text generation was only step one. If you’re building autonomous local AI agents, real time voice assistants, or edge workflows, instant low latency audio is the missing piece. Thanks to PR #26254, Alibaba’s state of the art Qwen3 TTS model family is now natively supported directly inside the llama.cpp repository under the multimodal (mtmd) framework. No Python bloat. No massive PyTorch CUDA overhead. Just raw, hyper optimized C++ running GGUF voice weights. Here is why this native update is a massive deal for the open source local AI stack: # Multimodal Architecture (.gguf + mmproj) Qwen3-TTS splits the workload between the base language model backbone and a multimodal projection adapter. llama.cpp handles this using the llama-tts binary, mapping the text model alongside its --mmproj projector to process audio tokens seamlessly. # Zero Shot Voice Cloning in Seconds You don't need fine tuning or massive dataset training. Feed the C++ engine a single 5 to 10 second .wav audio sample using the --tts-speaker-file flag, and it accurately clones the exact timbre, tone, and accent on the fly. # Real World T4 GPU Benchmark & Resource FootprintRunning the 1.7B Base model in 8-bit quantization (Q8_0): - VRAM Footprint: ~7 GB peak VRAM during active zero-shot cloning. - Audio Quality: Studio grade, natural-sounding voice output in seconds. • - Execution: Direct execution via native compiled binaries or sub process calls. # Coming Next to llama-server (PR #26603) Beyond CLI execution, a native POST /tts HTTP endpoint is currently being added to llama-server, which will soon allow you to trigger voice generation directly via standard REST API requests! # quick note on Colab compilation: Because this code was merged into mainline very recently, pre-built third-party binaries haven't fully caught up yet. Compiling llama-tts directly from source on Google Colab's free CPU instance can take about 1 hour (or ~1-2 minutes if targeting single GPU arch like -DCMAKE_CUDA_ARCHITECTURES=75). Be patient during the build step, or compile it locally on your own rig for instant execution! To test this out yourself, I built a zero config Google Colab notebook that compiles llama.cpp, downloads the Q8_0 GGUF files from HuggingFace, and spins up an interactive Gradio Studio UI so you can record/upload 3 second clips and clone voices in real time. Stop sleeping on native C++ audio. The era of bulky Python audio pipelines is officially over. Links to the free Google Colab notebook and the official ggml org GGUF HuggingFace model repository are in the replies below! available in q4 and q8 both variants, 1 GB and 1.85 GBs respectively (requires additional ~500MB mmproj gguf) Are you building local voice agents yet? What does your current audio stack look like? Drop your setups below!

Alok

47,881 просмотров • 1 месяц назад

🚨 WARNING: SOMETHING EXTREMELY BAD IS COMING ON MONDAY... Four things are breaking at the same time. Interest rate hikes by December is almost confirmed now. Not pauses. Not holds. HIKES. Everything priced on the assumption that cheap money returns just got repriced from scratch. Every leveraged position, every rate-sensitive asset, every model built on a pivot that isn't coming. Japan officially entered yen intervention, that sounds technical. Here's what it means in practice. Japan is burning through reserves defending a currency that's been collapsing for months. And to fund that defense they sell what they own. US Treasuries, at scale, Into a bond market that's already under pressure from every other direction. China hasn't stopped, while everyone was watching Iran and the Fed, China has been quietly and consistently dumping US Treasury holdings for weeks. No announcement, no fanfare. Just steady selling that shows up in the data for anyone paying attention. And the AI rally is dying in real time. The momentum that carried Nvidia, Microsoft, Google, and Meta to valuations that made no fundamental sense is reversing. Funds that loaded up on the AI narrative at the top are now the ones selling into every bounce. The multiple that drove three years of outperformance doesn't survive a rate hike cycle and a liquidity crisis simultaneously. Now put all four together. Rates going up, liquidity going out. The world's two largest foreign Treasury holders selling simultaneously. And the primary growth narrative of the last three years losing believers by the session. When liquidity disappears across multiple layers of the financial system at once markets don't correct. They don't dip, they don't give you a clean exit. They crash fast and they crash hard. Stocks. Bonds. Metals. Crypto. There is no rotation trade that works when everything is being sold to cover everything else. There is no safe haven when the people who need cash are selling whatever has a bid. Insiders are already out, funds are already cutting. The rotation happened this week while retail was still debating whether to buy the dip. By the time Monday's open confirms what's coming it's already too late to position. This sounds SCARY, but I will keep you updated on everything here When I rotate money, I will post my moves here so my FOLLOWERS can SAVE their money Many will regret not following me earlier...
0:15

🚨 WARNING: SOMETHING EXTREMELY BAD IS COMING ON MONDAY... Four things are breaking at the same time. Interest rate hikes by December is almost confirmed now. Not pauses. Not holds. HIKES. Everything priced on the assumption that cheap money returns just got repriced from scratch. Every leveraged position, every rate-sensitive asset, every model built on a pivot that isn't coming. Japan officially entered yen intervention, that sounds technical. Here's what it means in practice. Japan is burning through reserves defending a currency that's been collapsing for months. And to fund that defense they sell what they own. US Treasuries, at scale, Into a bond market that's already under pressure from every other direction. China hasn't stopped, while everyone was watching Iran and the Fed, China has been quietly and consistently dumping US Treasury holdings for weeks. No announcement, no fanfare. Just steady selling that shows up in the data for anyone paying attention. And the AI rally is dying in real time. The momentum that carried Nvidia, Microsoft, Google, and Meta to valuations that made no fundamental sense is reversing. Funds that loaded up on the AI narrative at the top are now the ones selling into every bounce. The multiple that drove three years of outperformance doesn't survive a rate hike cycle and a liquidity crisis simultaneously. Now put all four together. Rates going up, liquidity going out. The world's two largest foreign Treasury holders selling simultaneously. And the primary growth narrative of the last three years losing believers by the session. When liquidity disappears across multiple layers of the financial system at once markets don't correct. They don't dip, they don't give you a clean exit. They crash fast and they crash hard. Stocks. Bonds. Metals. Crypto. There is no rotation trade that works when everything is being sold to cover everything else. There is no safe haven when the people who need cash are selling whatever has a bid. Insiders are already out, funds are already cutting. The rotation happened this week while retail was still debating whether to buy the dip. By the time Monday's open confirms what's coming it's already too late to position. This sounds SCARY, but I will keep you updated on everything here When I rotate money, I will post my moves here so my FOLLOWERS can SAVE their money Many will regret not following me earlier...

ᴛʀᴀᴄᴇʀ

169,503 просмотров • 2 месяцев назад

I spend my days explaining to teams why a 770-billion-parameter open-weight model will never fit into their infrastructure. This week, I asked it to code a complete mobile game from a single prompt. The concept is one everyone has probably played before. A hole moving through an open-plan office, viewed from above, swallowing everything in its path. You start tiny, only able to swallow pens and cups. You grow, moving on to keyboards and plants, then chairs and printers, then desks and vending machines. Eventually, you swallow the entire meeting room. 60 seconds on the clock. One HTML file, zero external libraries, zero errors on launch. The video shows the generation and then the actual gameplay. What broke is more instructive than what worked. The structure came out right on the first try: fixed-timestep loop, spatial grid for collisions, tier system, spring camera. The balancing and rendering, not so much. The first version scored 70 points in 13 seconds with a tier threshold at 500, and drew colorful circles and triangles instead of furniture. I had to give it numbers and exact recipes. Speed: 640 pixels per second. Radii: 30, 58, 96, 150, 225. Tier thresholds: 150, 550, 1500, 3500. And for every object, a pixel-perfect drawing recipe. Once I gave it that, it followed the instructions exactly. A model that doesn't execute its own code won't tune itself. But it will execute, down to the exact numbers, what you tell it to build. The model is Hy4 preview, released by Tencent Hunyuan on August 28. 770 billion parameters in total, but 49 billion active per token. And that second number is what determines your serving bill. Native 1M context. Apache 2.0 license. vLLM and SGLang supported from day one, with an official FP8 checkpoint. Text-only preview. The part that matters for deployment is the compression. Tencent describes it as seven times smaller with almost no loss. GGUF builds use mixed per-layer quantization, where calibration data determines the bit width layer by layer. Some layers go as low as 1.31 bits, while others go up to 2.06 bits, averaging 2.38 bits per weight. The model drops from 1.5 TB in BF16 to 213.66 GiB while, according to their measurements, staying in the same performance range on real-world tasks. On this build, they report 204 tokens/s in prefill and 20 tokens/s in decoding, measured on an 8-GPU node. Their numbers, not mine. Their blind evaluation scores 2.99 out of 4 across 203 engineering tasks rated by 163 experts. Ahead of Kimi K3 at 2.94 and GLM-5.3 at 2.92. Their numbers too. My run went through the official hosted studio, not a local build, so I’m not claiming to have benchmarked the compressed GGUF myself. Two honest caveats. None of these builds run on standard llama.cpp. The hyv4 architecture isn't upstream yet, so patches are required. And 214 GiB of resident weights is still a server, not your laptop. This is a preview, and Tencent explicitly asks users to break it and report what fails. So here's my contribution.
0:19

I spend my days explaining to teams why a 770-billion-parameter open-weight model will never fit into their infrastructure. This week, I asked it to code a complete mobile game from a single prompt. The concept is one everyone has probably played before. A hole moving through an open-plan office, viewed from above, swallowing everything in its path. You start tiny, only able to swallow pens and cups. You grow, moving on to keyboards and plants, then chairs and printers, then desks and vending machines. Eventually, you swallow the entire meeting room. 60 seconds on the clock. One HTML file, zero external libraries, zero errors on launch. The video shows the generation and then the actual gameplay. What broke is more instructive than what worked. The structure came out right on the first try: fixed-timestep loop, spatial grid for collisions, tier system, spring camera. The balancing and rendering, not so much. The first version scored 70 points in 13 seconds with a tier threshold at 500, and drew colorful circles and triangles instead of furniture. I had to give it numbers and exact recipes. Speed: 640 pixels per second. Radii: 30, 58, 96, 150, 225. Tier thresholds: 150, 550, 1500, 3500. And for every object, a pixel-perfect drawing recipe. Once I gave it that, it followed the instructions exactly. A model that doesn't execute its own code won't tune itself. But it will execute, down to the exact numbers, what you tell it to build. The model is Hy4 preview, released by Tencent Hunyuan on August 28. 770 billion parameters in total, but 49 billion active per token. And that second number is what determines your serving bill. Native 1M context. Apache 2.0 license. vLLM and SGLang supported from day one, with an official FP8 checkpoint. Text-only preview. The part that matters for deployment is the compression. Tencent describes it as seven times smaller with almost no loss. GGUF builds use mixed per-layer quantization, where calibration data determines the bit width layer by layer. Some layers go as low as 1.31 bits, while others go up to 2.06 bits, averaging 2.38 bits per weight. The model drops from 1.5 TB in BF16 to 213.66 GiB while, according to their measurements, staying in the same performance range on real-world tasks. On this build, they report 204 tokens/s in prefill and 20 tokens/s in decoding, measured on an 8-GPU node. Their numbers, not mine. Their blind evaluation scores 2.99 out of 4 across 203 engineering tasks rated by 163 experts. Ahead of Kimi K3 at 2.94 and GLM-5.3 at 2.92. Their numbers too. My run went through the official hosted studio, not a local build, so I’m not claiming to have benchmarked the compressed GGUF myself. Two honest caveats. None of these builds run on standard llama.cpp. The hyv4 architecture isn't upstream yet, so patches are required. And 214 GiB of resident weights is still a server, not your laptop. This is a preview, and Tencent explicitly asks users to break it and report what fails. So here's my contribution.

Alexa Benchmark

16,315 просмотров • 21 дней назад

Anthropic Just Shot Itself in the Foot Anthropic launched Fable 5 and Mythos 5, then watched the US government shut them down three days later. The same government their CEO Dario Amodei has been begging for years to regulate AI harder. Now he got exactly what he asked for. This is straight-up leadership failure. Dario spent all that time pushing for rules and oversight. Those rules just killed his flagship models overnight. Customers in the middle of builds got cut off. Security teams using the models to find vulnerabilities suddenly had nothing. The company tried to call it a narrow export control thing over a jailbreak, but nobody is buying that spin. I helped move big clients off Anthropic the same night. One account alone was worth millions a month. They switched to local open-source models and they are not coming back. This is going to leave permanent damage. Customer exodus, key people leaving, and their IPO plans looking dead by the end of summer. This hurts US AI competitiveness and national security work. It pushes people toward open-source options, including ones from China. All because Anthropic positioned itself as the “safe and responsible” company that wanted government help. Now that help just flipped the off switch on their best stuff. Let’s run through Dario’s greatest hits of fear-mongering and delay tactics, because the pattern is ridiculous: • Back in 2019 at OpenAI, he helped push the call that GPT-2 was too dangerous to release fully. The world needed time to prepare, they said. It eventually came out anyway, and here we are. Did the sky fall? • He left OpenAI to start Anthropic, preaching “safe” AI with heavy guardrails, Constitutional AI, and all the rest. • Then came the endless public pleas for pauses, regulations, government audits, FAA-style oversight, export controls, and the power to block deployments. Essay after essay warning about risks while his company kept scaling. • Right up to recent weeks, Dario was still out there calling for stronger rules, pauses on frontier models, and giving governments the kill switch. And now? His own Mythos-class models get yanked by the bureaucracy he helped invite in. The clown show is complete. This is ridiculous. In two years, everyone will have Mythos-class AI — or better — running in their pocket, on their devices, with no guardrails, no corporate nanny filters, and no remote kill switch. Local, open-source, unstoppable. History is going to laugh at this entire episode: the CEO who spent years slowing everyone down only to watch his own company self-destruct by inviting the regulators to the party. Dario wanted regulation. He got it. The rest of the industry gets the lesson: inviting the state into your tech is a fast way to lose control of it. Centralized models like this are too fragile. Open-source and local alternatives just picked up a lot more users who will never trust a company like Anthropic again. This whole mess was completely avoidable. Hubris dressed up as safety advocacy. Now the bill is due.
0:10

Anthropic Just Shot Itself in the Foot Anthropic launched Fable 5 and Mythos 5, then watched the US government shut them down three days later. The same government their CEO Dario Amodei has been begging for years to regulate AI harder. Now he got exactly what he asked for. This is straight-up leadership failure. Dario spent all that time pushing for rules and oversight. Those rules just killed his flagship models overnight. Customers in the middle of builds got cut off. Security teams using the models to find vulnerabilities suddenly had nothing. The company tried to call it a narrow export control thing over a jailbreak, but nobody is buying that spin. I helped move big clients off Anthropic the same night. One account alone was worth millions a month. They switched to local open-source models and they are not coming back. This is going to leave permanent damage. Customer exodus, key people leaving, and their IPO plans looking dead by the end of summer. This hurts US AI competitiveness and national security work. It pushes people toward open-source options, including ones from China. All because Anthropic positioned itself as the “safe and responsible” company that wanted government help. Now that help just flipped the off switch on their best stuff. Let’s run through Dario’s greatest hits of fear-mongering and delay tactics, because the pattern is ridiculous: • Back in 2019 at OpenAI, he helped push the call that GPT-2 was too dangerous to release fully. The world needed time to prepare, they said. It eventually came out anyway, and here we are. Did the sky fall? • He left OpenAI to start Anthropic, preaching “safe” AI with heavy guardrails, Constitutional AI, and all the rest. • Then came the endless public pleas for pauses, regulations, government audits, FAA-style oversight, export controls, and the power to block deployments. Essay after essay warning about risks while his company kept scaling. • Right up to recent weeks, Dario was still out there calling for stronger rules, pauses on frontier models, and giving governments the kill switch. And now? His own Mythos-class models get yanked by the bureaucracy he helped invite in. The clown show is complete. This is ridiculous. In two years, everyone will have Mythos-class AI — or better — running in their pocket, on their devices, with no guardrails, no corporate nanny filters, and no remote kill switch. Local, open-source, unstoppable. History is going to laugh at this entire episode: the CEO who spent years slowing everyone down only to watch his own company self-destruct by inviting the regulators to the party. Dario wanted regulation. He got it. The rest of the industry gets the lesson: inviting the state into your tech is a fast way to lose control of it. Centralized models like this are too fragile. Open-source and local alternatives just picked up a lot more users who will never trust a company like Anthropic again. This whole mess was completely avoidable. Hubris dressed up as safety advocacy. Now the bill is due.

Brian Roemmele

141,854 просмотров • 3 месяцев назад

I just built a Meta ad policy checker in Claude Code that catches rejections BEFORE Meta does 🤯 Drop in your ad copy → it pulls Meta's LIVE Advertising Standards, checks every line against the actual policy text, and hands each ad a verdict: Cleared for launch, Fix before launch, or Grounded. All inside Claude Code. Perfect for media buyers and DTC brands who've had ads bounced — or an account restricted — and never got a straight answer why. If you're finding out about policy problems only after the rejection email, resubmitting the same ad and praying, losing days of delivery while the appeal sits in review, and every bounce quietly teaches Meta to trust your account a little less... This runs the review before Meta ever sees the ad: → Drop in your ad copy (one ad or a whole batch) → It reads each ad and figures out which of Meta's policies apply → Scrapes the live policy pages from Meta's Transparency Center → Flags the exact phrase that violates, with Meta's own policy quoted next to it → Rewrites the risky lines so the message survives but the violation doesn't → Renders a dashboard: every ad, every finding, every fix in one place No guessing which word killed the ad. No resubmit-and-pray loops. No stacking rejections on your account history. What you get: → A verdict on every ad before you spend a dollar → The violating phrase + the policy citation, side by side → Rewrites that keep the selling intent → A report you can hand straight to your team or client Built 100% in Claude Code. No API keys, no Meta login. I'm giving away the complete Claude skill file. Want the skill for free? > Like this post > Comment "META" And I'll send it over (must be following so I can DM)
0:18

I just built a Meta ad policy checker in Claude Code that catches rejections BEFORE Meta does 🤯 Drop in your ad copy → it pulls Meta's LIVE Advertising Standards, checks every line against the actual policy text, and hands each ad a verdict: Cleared for launch, Fix before launch, or Grounded. All inside Claude Code. Perfect for media buyers and DTC brands who've had ads bounced — or an account restricted — and never got a straight answer why. If you're finding out about policy problems only after the rejection email, resubmitting the same ad and praying, losing days of delivery while the appeal sits in review, and every bounce quietly teaches Meta to trust your account a little less... This runs the review before Meta ever sees the ad: → Drop in your ad copy (one ad or a whole batch) → It reads each ad and figures out which of Meta's policies apply → Scrapes the live policy pages from Meta's Transparency Center → Flags the exact phrase that violates, with Meta's own policy quoted next to it → Rewrites the risky lines so the message survives but the violation doesn't → Renders a dashboard: every ad, every finding, every fix in one place No guessing which word killed the ad. No resubmit-and-pray loops. No stacking rejections on your account history. What you get: → A verdict on every ad before you spend a dollar → The violating phrase + the policy citation, side by side → Rewrites that keep the selling intent → A report you can hand straight to your team or client Built 100% in Claude Code. No API keys, no Meta login. I'm giving away the complete Claude skill file. Want the skill for free? > Like this post > Comment "META" And I'll send it over (must be following so I can DM)

Mike Futia

17,526 просмотров • 2 месяцев назад

Why is the market selling off today? (Save this). The semi selloff right now is being driven by a mix of macro fear, profit taking and investors questioning how quickly all of this AI spending will actually pay off, not because demand for AI infrastructure suddenly disappeared. The market is basically trading this chain reaction, the ongoing US Iran escalation pushes oil higher, higher oil keeps inflation elevated, sticky inflation keeps Treasury yields high and that increases the risk of the Fed staying hawkish or even hiking again. That is a terrible setup for semis because many of these companies are valued on the massive earnings investors expect them to generate years from now. When yields rise, those future earnings become worth less today which is why the highest multiple AI and semiconductor names usually get hit first. (I don't think there will be a hike this year). This is also why everything is moving together right now. Nvidia, Micron, Nebius, SanDisk, Broadcom and Applied Optoelectronics are all completely different businesses, but institutions are not separating memory, networking, optics, compute and cloud infrastructure at the moment. They are reducing exposure to the entire AI trade, taking profits in the names that have already run the most and moving into a more defensive position potentially ahead of the Fed. There is also growing pressure around hyperscaler capex. Microsoft, Meta, Amazon and Google are still spending enormous amounts on GPUs, data centers, networking and power but the market is starting to ask when all of that spending will actually turn into revenue and free cash flow. Investors are no longer satisfied with hearing that AI capex is growing. They want proof that the returns are arriving fast enough to justify the valuations already priced into the entire AI ecosystem. That creates a weird situation where hyperscaler capex can continue rising while semiconductor stocks still fall. The market is not asking whether AI spending is growing anymore but rather asking whether it is growing fast enough to beat the expectations already baked into these stocks. Crowded positioning is another major factor. Semis and AI infrastructure stocks have been some of the biggest winners in the market so institutions are sitting on huge profits and many funds own the exact same names. When macro risk increases, investors usually sell the most liquid winners first. That does not mean demand for memory, optics or custom chips suddenly collapsed but rather means investors are locking in gains and reducing risk. Tariffs add another layer because even when they are not directly placed on chips, they can still raise the cost of servers, electrical equipment, cooling systems, construction materials and the overall data center buildout. That makes AI infrastructure more expensive while also adding another source of inflation. Then you have Jensen Huang’s letter to the White House this morning about open weight AI models, which I think is one of the most important long term developments here. Nvidia, Meta, Microsoft, Palantir and several other companies are pushing Washington not to place broad restrictions on open weight AI. OpenAI and Anthropic were notably absent because open models are much more of a threat to their business models. OpenAI and Anthropic benefit from a world where a few closed frontier labs control the best models and companies have to pay them through subscriptions and APIs. Open weight models weaken that advantage because businesses can download a model, customize it for their own use and run it on their own infrastructure or through a neocloud. That is bad for OpenAI and Anthropic because it puts pressure on pricing, margins and the idea that they will control the intelligence layer of the economy but it is very good for the AI ecosystem as a whole over the long run. But the question is what does this mean for all the OpenAI and Anthropic commitments? so that's adding to the fear as well. But with that being said open models make AI cheaper and more accessible. Instead of AI being controlled by a few giant labs, thousands of startups, universities, governments and regular businesses can deploy models themselves. That spreads AI adoption across the entire economy and creates a much larger infrastructure opportunity and that is exactly why Jensen cares. Nvidia does not need OpenAI or Anthropic to win. Nvidia just needs more people using AI. Whether the model comes from OpenAI, Anthropic, Meta, Mistral, Kimi or some startup nobody has heard of yet, it still needs GPUs, memory, networking, data centers and electricity. So open weight AI could actually weaken the model companies while making the infrastructure layer much bigger. More open models mean more companies running inference. More inference means more GPUs. More GPUs mean more HBM, optical transceivers, switches, data centers and power. That is bullish for Nvidia Nebius, Micron, Broadcom , Marvell and Applied Optoelectronics over the long run. So my take is that the current semi selloff is being driven mostly by macro uncertainty, higher oil, rising yields, Fed fears, tariffs, crowded positioning and questions around the return on hyperscaler capex. The underlying AI infrastructure thesis has not suddenly broken. We are not broadly seeing hyperscalers cancel GPU orders, slash capex, abandon data center projects or report that AI demand has collapsed. What has changed is the valuation investors are willing to pay while the macro environment remains unstable. The market is lowering the price it is willing to pay for semiconductor growth but is not necessarily saying that growth is gone. And while Jensen’s open weight push may be bad for OpenAI and Anthropic, it could be one of the best things possible for the AI ecosystem over the long run because it creates more models, more developers, more competition and ultimately much more demand for the infrastructure underneath all of it. Nothing about the AI thesis has changed for me, so I will be going shopping and taking advantage of this sale while the market is selling everything together. I am an analyst at Milk Road Pro, and if you want to see exactly what I am buying, you can join for just $1 using the link below.
0:11

Why is the market selling off today? (Save this). The semi selloff right now is being driven by a mix of macro fear, profit taking and investors questioning how quickly all of this AI spending will actually pay off, not because demand for AI infrastructure suddenly disappeared. The market is basically trading this chain reaction, the ongoing US Iran escalation pushes oil higher, higher oil keeps inflation elevated, sticky inflation keeps Treasury yields high and that increases the risk of the Fed staying hawkish or even hiking again. That is a terrible setup for semis because many of these companies are valued on the massive earnings investors expect them to generate years from now. When yields rise, those future earnings become worth less today which is why the highest multiple AI and semiconductor names usually get hit first. (I don't think there will be a hike this year). This is also why everything is moving together right now. Nvidia, Micron, Nebius, SanDisk, Broadcom and Applied Optoelectronics are all completely different businesses, but institutions are not separating memory, networking, optics, compute and cloud infrastructure at the moment. They are reducing exposure to the entire AI trade, taking profits in the names that have already run the most and moving into a more defensive position potentially ahead of the Fed. There is also growing pressure around hyperscaler capex. Microsoft, Meta, Amazon and Google are still spending enormous amounts on GPUs, data centers, networking and power but the market is starting to ask when all of that spending will actually turn into revenue and free cash flow. Investors are no longer satisfied with hearing that AI capex is growing. They want proof that the returns are arriving fast enough to justify the valuations already priced into the entire AI ecosystem. That creates a weird situation where hyperscaler capex can continue rising while semiconductor stocks still fall. The market is not asking whether AI spending is growing anymore but rather asking whether it is growing fast enough to beat the expectations already baked into these stocks. Crowded positioning is another major factor. Semis and AI infrastructure stocks have been some of the biggest winners in the market so institutions are sitting on huge profits and many funds own the exact same names. When macro risk increases, investors usually sell the most liquid winners first. That does not mean demand for memory, optics or custom chips suddenly collapsed but rather means investors are locking in gains and reducing risk. Tariffs add another layer because even when they are not directly placed on chips, they can still raise the cost of servers, electrical equipment, cooling systems, construction materials and the overall data center buildout. That makes AI infrastructure more expensive while also adding another source of inflation. Then you have Jensen Huang’s letter to the White House this morning about open weight AI models, which I think is one of the most important long term developments here. Nvidia, Meta, Microsoft, Palantir and several other companies are pushing Washington not to place broad restrictions on open weight AI. OpenAI and Anthropic were notably absent because open models are much more of a threat to their business models. OpenAI and Anthropic benefit from a world where a few closed frontier labs control the best models and companies have to pay them through subscriptions and APIs. Open weight models weaken that advantage because businesses can download a model, customize it for their own use and run it on their own infrastructure or through a neocloud. That is bad for OpenAI and Anthropic because it puts pressure on pricing, margins and the idea that they will control the intelligence layer of the economy but it is very good for the AI ecosystem as a whole over the long run. But the question is what does this mean for all the OpenAI and Anthropic commitments? so that's adding to the fear as well. But with that being said open models make AI cheaper and more accessible. Instead of AI being controlled by a few giant labs, thousands of startups, universities, governments and regular businesses can deploy models themselves. That spreads AI adoption across the entire economy and creates a much larger infrastructure opportunity and that is exactly why Jensen cares. Nvidia does not need OpenAI or Anthropic to win. Nvidia just needs more people using AI. Whether the model comes from OpenAI, Anthropic, Meta, Mistral, Kimi or some startup nobody has heard of yet, it still needs GPUs, memory, networking, data centers and electricity. So open weight AI could actually weaken the model companies while making the infrastructure layer much bigger. More open models mean more companies running inference. More inference means more GPUs. More GPUs mean more HBM, optical transceivers, switches, data centers and power. That is bullish for Nvidia Nebius, Micron, Broadcom , Marvell and Applied Optoelectronics over the long run. So my take is that the current semi selloff is being driven mostly by macro uncertainty, higher oil, rising yields, Fed fears, tariffs, crowded positioning and questions around the return on hyperscaler capex. The underlying AI infrastructure thesis has not suddenly broken. We are not broadly seeing hyperscalers cancel GPU orders, slash capex, abandon data center projects or report that AI demand has collapsed. What has changed is the valuation investors are willing to pay while the macro environment remains unstable. The market is lowering the price it is willing to pay for semiconductor growth but is not necessarily saying that growth is gone. And while Jensen’s open weight push may be bad for OpenAI and Anthropic, it could be one of the best things possible for the AI ecosystem over the long run because it creates more models, more developers, more competition and ultimately much more demand for the infrastructure underneath all of it. Nothing about the AI thesis has changed for me, so I will be going shopping and taking advantage of this sale while the market is selling everything together. I am an analyst at Milk Road Pro, and if you want to see exactly what I am buying, you can join for just $1 using the link below.

Melvin

180,578 просмотров • 2 месяцев назад

vPay offshore accounts and physical cards have been getting field-tested IRL for a while now, and we’ll open them to the public as soon as we’re fully confident in the UX. But before offshore accounts go public, I want to address a few points: Some might point out that - vPay isn’t the first crypto card - vPay doesn’t have the lowest fees - So why choose vPay instead of the Coinbase 🛡️ Card or MetaMask 🦊 Card or KAST or or Tria, or any of the other big names? Now to address: Privacy | The biggest differentiator that sets vPay completely apart is Private Banking. The majority of the crypto card providers on the market use Rain infra. Even if you’ve never heard of them, that's what your favorite "NeoBank" uses. And due to their legal jurisdictions, they will report your finances to authorities since they're CRS and FACTA compliant. We are not. As an OmniBank, we work with different banking partners, and although KYC is required to use our services, our offshore banks are non-CRS and non-FACTA. Tax reporting is the responsibility and choice of the user. Offshore Accounts vs Physical Cards | I've tried to highlight this a few times so far. vPay has 3 offerings on the banking side of things. Virtual cards - live now. Physical cards - coming Q1 2026. The first two are similar to what everyone else on the market offers. The offshore accounts are not. which are coming this week. They allow unlimited spending, ATM withdrawals, and international SWIFT transfers, which very few “Neobanks” provide. Offshore accounts are coming this week. Self-Custody | We're not 100% non-custodial yet, as that is near impossible at the moment but it's something we're working towards. And we try to keep the users' self-custodial wallets in the loop as much as possible for maximum control. Those who have tried the vPay app know that almost every move asks for permission from their wallet, and we always encourage users to keep their funds in their non-custodial wallets until the very last moment, since our top-ups usually only take seconds to a minute to process. Fees | All of the card providers mentioned above either raised millions from VCs or in presales or have a huge org backing them. We have neither. vPay was self-funded and community-owned since day 1, launched under Virtuals Protocol Genesis V1 launch model, an objectively bad launch model and hugely unfavorable toward project teams. So even though vPay has been generating revenue and profitable from early on, we do not have the luxury of offering 0% fees yet, since they're mostly a marketing gimmick paid for by millions in VC money and not a sustainable business model for early-stage companies. What we're working towards instead, is true co-ownership of vPay and revenue-share with users. OmniBank vs NeoBank | I’m not a fan of the term “NeoBank.” It implies just a bank, but make it crypto. That’s not vPay. Our goals have always been clear: A) Anything and everything users need to do with their money and assets, both Web2 and Web3, all in one hub. Powered by a constellation of partner agents. The cards and the bank accounts are just the foundation. B) To eventually build independent financial rails for crypto and decouple from the chokehold of Visa/Mastercard. vLink is the first step toward this vision. This turned out to be a rather long tweet, but context matters. Questions and feedback welcome in replies or DMs. See you all with your vPay vCards very soon.
0:08

vPay offshore accounts and physical cards have been getting field-tested IRL for a while now, and we’ll open them to the public as soon as we’re fully confident in the UX. But before offshore accounts go public, I want to address a few points: Some might point out that - vPay isn’t the first crypto card - vPay doesn’t have the lowest fees - So why choose vPay instead of the Coinbase 🛡️ Card or MetaMask 🦊 Card or KAST or or Tria, or any of the other big names? Now to address: Privacy | The biggest differentiator that sets vPay completely apart is Private Banking. The majority of the crypto card providers on the market use Rain infra. Even if you’ve never heard of them, that's what your favorite "NeoBank" uses. And due to their legal jurisdictions, they will report your finances to authorities since they're CRS and FACTA compliant. We are not. As an OmniBank, we work with different banking partners, and although KYC is required to use our services, our offshore banks are non-CRS and non-FACTA. Tax reporting is the responsibility and choice of the user. Offshore Accounts vs Physical Cards | I've tried to highlight this a few times so far. vPay has 3 offerings on the banking side of things. Virtual cards - live now. Physical cards - coming Q1 2026. The first two are similar to what everyone else on the market offers. The offshore accounts are not. which are coming this week. They allow unlimited spending, ATM withdrawals, and international SWIFT transfers, which very few “Neobanks” provide. Offshore accounts are coming this week. Self-Custody | We're not 100% non-custodial yet, as that is near impossible at the moment but it's something we're working towards. And we try to keep the users' self-custodial wallets in the loop as much as possible for maximum control. Those who have tried the vPay app know that almost every move asks for permission from their wallet, and we always encourage users to keep their funds in their non-custodial wallets until the very last moment, since our top-ups usually only take seconds to a minute to process. Fees | All of the card providers mentioned above either raised millions from VCs or in presales or have a huge org backing them. We have neither. vPay was self-funded and community-owned since day 1, launched under Virtuals Protocol Genesis V1 launch model, an objectively bad launch model and hugely unfavorable toward project teams. So even though vPay has been generating revenue and profitable from early on, we do not have the luxury of offering 0% fees yet, since they're mostly a marketing gimmick paid for by millions in VC money and not a sustainable business model for early-stage companies. What we're working towards instead, is true co-ownership of vPay and revenue-share with users. OmniBank vs NeoBank | I’m not a fan of the term “NeoBank.” It implies just a bank, but make it crypto. That’s not vPay. Our goals have always been clear: A) Anything and everything users need to do with their money and assets, both Web2 and Web3, all in one hub. Powered by a constellation of partner agents. The cards and the bank accounts are just the foundation. B) To eventually build independent financial rails for crypto and decouple from the chokehold of Visa/Mastercard. vLink is the first step toward this vision. This turned out to be a rather long tweet, but context matters. Questions and feedback welcome in replies or DMs. See you all with your vPay vCards very soon.

The Dude

20,558 просмотров • 9 месяцев назад

✨ I open sourced my first Chrome extension 🚀 SuperLevels I vibe coded it to replace all my Chrome extensions that are increasingly being bought up by spyware and malware companies who sell your data or worse hack your accounts and steal your stuff/money/data, which I'd call one of the top security risks right now For example: Chrome extensions can read your cookies or localStorage data, including session tokens, then login to your web or email accounts and hack you, they can inject code into any site to pull data form any site you browse, then break into your crypto accounts, drain your wallets, and selling your browsing history to ad companies, but that'd actually be the most favorable thing to happen of all these! Chrome extensions are just very very very unsafe So I coded my own, that I can trust because I made it, and I can read the source code: my extension is called 🚀SuperLevels and has all the features that the Chrome extensions I used to use have but all built into one safe one The cool thing is it's 100% open source and free, and you can audit the code first with AI yourself before installing it, and then if you do install it, customize it to your liking again with AI It has these features that improve my daily workflow while browsing the web: 🚮 Tab Cleaner Automatically closes inactive tabs after a configurable timeout (default: 5 minutes). Set excluded hosts to keep important tabs alive. View and re-open recently closed tabs. 🍪 Cookie Editor Full cookie manager for the current site. View, edit, add, and delete cookies. Export cookies as JSON. Expand any cookie to see and modify all fields including domain, path, SameSite, secure, and httpOnly flags. 🔀 Redirect Tracer See every redirect hop your browser took to reach the current page. Shows status codes (301, 302, 307, etc.) with a visual chain. Copy the full redirect chain to clipboard. 🌙 Dark Mode Instant dark mode for any website using CSS filter inversion. Adjustable brightness. Toggle per-site or globally. Images and videos are automatically re-inverted so they look normal. 𝕏 X Dim Mode Custom dim theme for X/Twitter with 7 color palettes: Dim, Slate, Jade, Plum, Dusk, Ember, or a custom hue. Live preview in the popup. ⚡ JS Toggle Disable JavaScript per-site with one click. Useful for debugging, reading articles without popups, or testing progressive enhancement. Page reloads automatically. 🚫 GDPR Cookie Consent Dismisser Auto-hides and auto-clicks cookie consent banners. Supports OneTrust, CookieBot, Didomi, Quantcast, GDPR plugins, and dozens more frameworks. Toggle off if a site breaks. 🎨 Live CSS Editor Write custom CSS for any website, applied in real-time as you type. Saved per-domain. Supports tab key for indentation. 📺 YouTube Unhook Removes YouTube distractions: no homepage feed, no sidebar suggestions, no end screen overlays, no Shorts. Search still works — just no algorithmic recommendations. 🎵 Music Recognizer Shazam-like music identification for any tab. Captures 10 seconds of audio and identifies the song via ACRCloud (free signup, bring your own API key). Results link to YouTube. History of recognized songs. 🖼 Picture-in-Picture Pop the largest video on the current tab into a floating PiP window with one click. 🗺 Google Maps Links Re-adds clickable Maps links and map preview cards to Google Search results. 🖼 View Image Adds a "View Image" button back to Google Images, linking directly to the full-size original image. {} JSON Formatter Auto-detects pure JSON response pages and formats them with syntax highlighting, collapsible sections, and a dark theme. Copy or view raw with one click. Never triggers on regular HTML pages.
0:17

✨ I open sourced my first Chrome extension 🚀 SuperLevels I vibe coded it to replace all my Chrome extensions that are increasingly being bought up by spyware and malware companies who sell your data or worse hack your accounts and steal your stuff/money/data, which I'd call one of the top security risks right now For example: Chrome extensions can read your cookies or localStorage data, including session tokens, then login to your web or email accounts and hack you, they can inject code into any site to pull data form any site you browse, then break into your crypto accounts, drain your wallets, and selling your browsing history to ad companies, but that'd actually be the most favorable thing to happen of all these! Chrome extensions are just very very very unsafe So I coded my own, that I can trust because I made it, and I can read the source code: my extension is called 🚀SuperLevels and has all the features that the Chrome extensions I used to use have but all built into one safe one The cool thing is it's 100% open source and free, and you can audit the code first with AI yourself before installing it, and then if you do install it, customize it to your liking again with AI It has these features that improve my daily workflow while browsing the web: 🚮 Tab Cleaner Automatically closes inactive tabs after a configurable timeout (default: 5 minutes). Set excluded hosts to keep important tabs alive. View and re-open recently closed tabs. 🍪 Cookie Editor Full cookie manager for the current site. View, edit, add, and delete cookies. Export cookies as JSON. Expand any cookie to see and modify all fields including domain, path, SameSite, secure, and httpOnly flags. 🔀 Redirect Tracer See every redirect hop your browser took to reach the current page. Shows status codes (301, 302, 307, etc.) with a visual chain. Copy the full redirect chain to clipboard. 🌙 Dark Mode Instant dark mode for any website using CSS filter inversion. Adjustable brightness. Toggle per-site or globally. Images and videos are automatically re-inverted so they look normal. 𝕏 X Dim Mode Custom dim theme for X/Twitter with 7 color palettes: Dim, Slate, Jade, Plum, Dusk, Ember, or a custom hue. Live preview in the popup. ⚡ JS Toggle Disable JavaScript per-site with one click. Useful for debugging, reading articles without popups, or testing progressive enhancement. Page reloads automatically. 🚫 GDPR Cookie Consent Dismisser Auto-hides and auto-clicks cookie consent banners. Supports OneTrust, CookieBot, Didomi, Quantcast, GDPR plugins, and dozens more frameworks. Toggle off if a site breaks. 🎨 Live CSS Editor Write custom CSS for any website, applied in real-time as you type. Saved per-domain. Supports tab key for indentation. 📺 YouTube Unhook Removes YouTube distractions: no homepage feed, no sidebar suggestions, no end screen overlays, no Shorts. Search still works — just no algorithmic recommendations. 🎵 Music Recognizer Shazam-like music identification for any tab. Captures 10 seconds of audio and identifies the song via ACRCloud (free signup, bring your own API key). Results link to YouTube. History of recognized songs. 🖼 Picture-in-Picture Pop the largest video on the current tab into a floating PiP window with one click. 🗺 Google Maps Links Re-adds clickable Maps links and map preview cards to Google Search results. 🖼 View Image Adds a "View Image" button back to Google Images, linking directly to the full-size original image. {} JSON Formatter Auto-detects pure JSON response pages and formats them with syntax highlighting, collapsible sections, and a dark theme. Copy or view raw with one click. Never triggers on regular HTML pages.

@levelsio

259,929 просмотров • 5 месяцев назад

This is my "feel the AGI" moment: I used GPT-5.6 Sol to train my own autocorrect model that outperforms GPT-5.6 Sol (wtf??) I have no ML background. I have no idea what I'm doing. I just kept pushing Sol until it spat out a SOTA model. And I spent $0. The motivation: Years of talking to AI have made me terrible at typing. Rather than fix my skill issue, I decided to throw more AI at it. My idea was: instead of autocorrect that interrupts my flow, I want to type fast with mistakes and have AI clean it up after. I wanted the smallest local model possible, for speed, for battery life, for science! So I decided to train my own. Inspired by Andrej Karpathy’s autoresearch, I ran Codex /goal with this setup: pick an experiment, try it, record the results to a doc, throw it out if it fails, and plan the next experiment without repeating failures. I gave a few examples that had to pass, tight latency targets, and let it run. Sol did some amazing things. First, it scanned benchmarks and shortlisted base models: Qwen 3.5, Gemma 4, Liquid LFM 2.5. It found a dataset on HuggingFace for typed text. Then it built a simulator for fingers striking a Mac keyboard, modeling the physical layout with a Gaussian distribution around each key. It simulated striking the wrong key, wrong order, fat-fingering, etc. With the models + data + simulator, it fine-tuned using MLX right on my MacBook. It had a working prototype within an hour! But accuracy was pretty poor. — Problem 1: Tokenization Sol read papers, ran tests, and identified that the tokenizer was the bottleneck. Tokenization makes typos hard for the model to see, so it memorizes mappings instead of using its language priors. Sol tried ByT5, Google’s tokenizer-free byte-level LLM. This made a big improvement, but the model is old and lacked the knowledge needed to reach Sol performance. Sol dug deeper and realized a tokenizer-free model isn’t needed; instead, it used T5Gemma, an encoder-decoder model. This can understand the input deeply before producing output, and furthermore, Sol could post-train the encoder to improve performance. This gave a much higher ceiling. — Problem 2: Loss function Now the model was correcting some typos perfectly, but ignoring most. Sol realized that standard cross-entropy loss was teaching the model to avoid edits, because the vast majority of characters in the training data were left unmodified. The fix was wild: Sol wrote a custom loss function that byte-aligns the source and target strings, uses a dynamic programming algorithm to compute the minimum edits between the two, then weights correct edits much higher than copies. After a lot of tuning, this dramatically improved accuracy. — Problem 3: Autoregression One failure mode remained: if the model made a mistake, it couldn’t backtrack. It could only predict the next token. Teaching it to “think” like a reasoning model would solve this, but would be far too slow. Sol found a beautiful solution: instead of greedily predicting the next token, beam search over all possibilities. This parallelizes the exploration instead of one linear chain-of-thought. At the end, choose the path with highest cumulative log probability. This worked great, but made the experience worse, since the user wouldn’t see progress until the whole search was done. To fix this, Sol made a clever observation: after each search step, the longest common prefix among surviving branches is guaranteed to appear in the final result, so it can be displayed immediately. As the search progresses, weaker paths are dropped and the prefix grows, so the user sees continuous progress. Sol built all this as a custom MLX pipeline that does the parallel decoding on the MacBook GPU, with just ~40ms TTFT. It’s crazy fast and entirely local. — Final eval (error reduction rate, higher is better): - Apple autocorrect: 49.66% - GPT-5.6 Luna: 82.47% - GPT-5.6 Terra: 87.64% - GPT-5.6 Sol: 90.56% - Our model (1.7B): 91.02% Final cost: - 1 quota reset (thanks Tibo) - $0 (And yes, I verified there's no cheating. In fact, we test words scrubbed from the training data to prove the model isn’t memorizing) There were a ton more details and tangents I could write about: contrastive learning, GRPO, DPO, dynamic masking, and more. Sol is a fascinating and creative model. It blew my mind so many times. Don’t let a lack of experience stop you: Sol makes AI experiments accessible to anyone!
0:12

This is my "feel the AGI" moment: I used GPT-5.6 Sol to train my own autocorrect model that outperforms GPT-5.6 Sol (wtf??) I have no ML background. I have no idea what I'm doing. I just kept pushing Sol until it spat out a SOTA model. And I spent $0. The motivation: Years of talking to AI have made me terrible at typing. Rather than fix my skill issue, I decided to throw more AI at it. My idea was: instead of autocorrect that interrupts my flow, I want to type fast with mistakes and have AI clean it up after. I wanted the smallest local model possible, for speed, for battery life, for science! So I decided to train my own. Inspired by Andrej Karpathy’s autoresearch, I ran Codex /goal with this setup: pick an experiment, try it, record the results to a doc, throw it out if it fails, and plan the next experiment without repeating failures. I gave a few examples that had to pass, tight latency targets, and let it run. Sol did some amazing things. First, it scanned benchmarks and shortlisted base models: Qwen 3.5, Gemma 4, Liquid LFM 2.5. It found a dataset on HuggingFace for typed text. Then it built a simulator for fingers striking a Mac keyboard, modeling the physical layout with a Gaussian distribution around each key. It simulated striking the wrong key, wrong order, fat-fingering, etc. With the models + data + simulator, it fine-tuned using MLX right on my MacBook. It had a working prototype within an hour! But accuracy was pretty poor. — Problem 1: Tokenization Sol read papers, ran tests, and identified that the tokenizer was the bottleneck. Tokenization makes typos hard for the model to see, so it memorizes mappings instead of using its language priors. Sol tried ByT5, Google’s tokenizer-free byte-level LLM. This made a big improvement, but the model is old and lacked the knowledge needed to reach Sol performance. Sol dug deeper and realized a tokenizer-free model isn’t needed; instead, it used T5Gemma, an encoder-decoder model. This can understand the input deeply before producing output, and furthermore, Sol could post-train the encoder to improve performance. This gave a much higher ceiling. — Problem 2: Loss function Now the model was correcting some typos perfectly, but ignoring most. Sol realized that standard cross-entropy loss was teaching the model to avoid edits, because the vast majority of characters in the training data were left unmodified. The fix was wild: Sol wrote a custom loss function that byte-aligns the source and target strings, uses a dynamic programming algorithm to compute the minimum edits between the two, then weights correct edits much higher than copies. After a lot of tuning, this dramatically improved accuracy. — Problem 3: Autoregression One failure mode remained: if the model made a mistake, it couldn’t backtrack. It could only predict the next token. Teaching it to “think” like a reasoning model would solve this, but would be far too slow. Sol found a beautiful solution: instead of greedily predicting the next token, beam search over all possibilities. This parallelizes the exploration instead of one linear chain-of-thought. At the end, choose the path with highest cumulative log probability. This worked great, but made the experience worse, since the user wouldn’t see progress until the whole search was done. To fix this, Sol made a clever observation: after each search step, the longest common prefix among surviving branches is guaranteed to appear in the final result, so it can be displayed immediately. As the search progresses, weaker paths are dropped and the prefix grows, so the user sees continuous progress. Sol built all this as a custom MLX pipeline that does the parallel decoding on the MacBook GPU, with just ~40ms TTFT. It’s crazy fast and entirely local. — Final eval (error reduction rate, higher is better): - Apple autocorrect: 49.66% - GPT-5.6 Luna: 82.47% - GPT-5.6 Terra: 87.64% - GPT-5.6 Sol: 90.56% - Our model (1.7B): 91.02% Final cost: - 1 quota reset (thanks Tibo) - $0 (And yes, I verified there's no cheating. In fact, we test words scrubbed from the training data to prove the model isn’t memorizing) There were a ton more details and tangents I could write about: contrastive learning, GRPO, DPO, dynamic masking, and more. Sol is a fascinating and creative model. It blew my mind so many times. Don’t let a lack of experience stop you: Sol makes AI experiments accessible to anyone!

Anshu

179,451 просмотров • 2 месяцев назад

While most of the market was selling off the Fed was raising rates and the CLARITY Act was failing in the Senate Zcash decided to do the exact opposite zcash:native was trading close to $400 at the end of June Today it’s around $1,485 That move looks completely irrational until you look at what happened before it The rally actually started with a disaster Zcash launched in 2016 with the same 21M supply limit and halving structure people already understood from Bitcoin The important difference is privacy Bitcoin shows the sender, receiver and amount. Zcash can hide all 3 when someone chooses to use a shielded address Privacy isn’t forced on everyone, it’s optional That detail matters because the entire current story comes back to whether people are actually choosing it In June, researchers disclosed an inflation vulnerability inside Orchard, Zcash’s main shielded pool The bug had reportedly existed unnoticed for roughly four years and, in theory could have allowed someone to create counterfeit ZEC without the public being able to detect it There was no evidence that anyone had exploited it But once the possibility existed, the market had to ask an uncomfortable question Is the 21M supply really 21M? Price collapsed, people pulled coins from the shielded pool and confidence disappeared almost overnight Then Ironwood arrived in July The compromised Orchard pool was permanently sealed and replaced with a new shielded pool Coins moving into the new system had to pass through a controlled migration process allowing the supply to be accounted for again A counterfeit coin couldn’t simply enter the new pool unnoticed The code was repaired but more importantly the supply became believable again That’s the first half of the current ZEC move: repair The second half is expansion Roughly 28% of circulating ZEC is now held inside shielded pools up from around 23% a year ago Shielded activity has also grown substantially People aren’t only talking about privacy, more of them are actually using it Who owns those shielded coins? The honest answer is that we don’t know And not knowing isn’t a flaw in Zcash, it’s the product You can see value moving from transparent addresses into a shielded pool but you can’t see who controls it or where it goes afterwards That’s why claims that “governments are secretly buying” or “insiders are hiding supply” should be treated carefully, the system doesn’t give us enough information to prove or disprove either story What we can measure is the transparent side Grayscale’s zcash:native now gives investors regulated exposure to ZEC through a US-listed ETF, that demand isn’t hidden, it sits inside a regulated KYC-compliant product So the lazy “this is all black money” explanation doesn’t really fit the data we can see The simpler explanation is that demand for financial privacy rises as surveillance and restrictions increase Europe has placed additional pressure on privacy coins under MiCA, some exchanges have restricted ZEC and other jurisdictions are treating privacy assets as a separate regulatory category Normally, regulatory pressure is supposed to reduce demand With privacy, it can also remind people why the product exists Zcash is changing in another way too For years it had one clear job: private payments Now Zcash Shielded Assets are opening the door to private tokens, collectibles and NFT experiments. ZecBit is working on shielded NFT infrastructure while other projects are experimenting with private ownership and sealed-bid auctions That ecosystem is still early, I wouldn’t pretend otherwise But for the first time Zcash is developing a culture layer on top of its privacy layer That gives people another reason to hold and use the network beyond sending money The timing of this rally is probably what makes it stand out most The Fed raised rates The CLARITY Act failed to advance Bitcoin briefly traded around $75K ZEC still climbed roughly 35% While everyone else was waiting for direction from the Senate and the Fed a network built around escaping financial surveillance barely cared Maybe that’s the point For me this move stands on two legs The first is repaired trust after Ironwood The second is growing utility around shielded assets and applications The metric I’ll keep watching isn’t a politician a headline or a mystery whale It’s the amount of ZEC entering and remaining inside shielded pools If that number keeps rising the privacy premium has real usage behind it If it stalls the rally becomes much easier to describe as speculation There is a fair bearish argument here too F2Pool co-founder Chun Wang has described the move as a narrative-driven short squeeze. He has a point privacy still isn’t the default setting and a near-vertical move always attracts leverage and momentum traders The chart is stretched so I’m not treating every green candle as a safe entry $1,400 is the first short-term pivot Below that, $1,250-$1,300 is the main demand area I want to see defended A clean break and hold above the recent high around $1,530 would open $1,650 first followed by a possible extension toward $1,800 Lose $1,100 and the structure starts looking much weaker September 30 is the next date on my calendar, features that aren’t ready for the coming NU7 upgrade by that deadline may be dropped from this release So yes, part of this move may be a squeeze But it’s difficult to call the entire rally empty when the network has repaired its biggest credibility problem shielded usage is growing and a new application layer is beginning to appear at the same time Zcash spent years selling one idea Privacy when you need it The market may finally be deciding that it needs it Not financial advice, just how I’m reading the data and the chart
0:12

While most of the market was selling off the Fed was raising rates and the CLARITY Act was failing in the Senate Zcash decided to do the exact opposite zcash:native was trading close to $400 at the end of June Today it’s around $1,485 That move looks completely irrational until you look at what happened before it The rally actually started with a disaster Zcash launched in 2016 with the same 21M supply limit and halving structure people already understood from Bitcoin The important difference is privacy Bitcoin shows the sender, receiver and amount. Zcash can hide all 3 when someone chooses to use a shielded address Privacy isn’t forced on everyone, it’s optional That detail matters because the entire current story comes back to whether people are actually choosing it In June, researchers disclosed an inflation vulnerability inside Orchard, Zcash’s main shielded pool The bug had reportedly existed unnoticed for roughly four years and, in theory could have allowed someone to create counterfeit ZEC without the public being able to detect it There was no evidence that anyone had exploited it But once the possibility existed, the market had to ask an uncomfortable question Is the 21M supply really 21M? Price collapsed, people pulled coins from the shielded pool and confidence disappeared almost overnight Then Ironwood arrived in July The compromised Orchard pool was permanently sealed and replaced with a new shielded pool Coins moving into the new system had to pass through a controlled migration process allowing the supply to be accounted for again A counterfeit coin couldn’t simply enter the new pool unnoticed The code was repaired but more importantly the supply became believable again That’s the first half of the current ZEC move: repair The second half is expansion Roughly 28% of circulating ZEC is now held inside shielded pools up from around 23% a year ago Shielded activity has also grown substantially People aren’t only talking about privacy, more of them are actually using it Who owns those shielded coins? The honest answer is that we don’t know And not knowing isn’t a flaw in Zcash, it’s the product You can see value moving from transparent addresses into a shielded pool but you can’t see who controls it or where it goes afterwards That’s why claims that “governments are secretly buying” or “insiders are hiding supply” should be treated carefully, the system doesn’t give us enough information to prove or disprove either story What we can measure is the transparent side Grayscale’s zcash:native now gives investors regulated exposure to ZEC through a US-listed ETF, that demand isn’t hidden, it sits inside a regulated KYC-compliant product So the lazy “this is all black money” explanation doesn’t really fit the data we can see The simpler explanation is that demand for financial privacy rises as surveillance and restrictions increase Europe has placed additional pressure on privacy coins under MiCA, some exchanges have restricted ZEC and other jurisdictions are treating privacy assets as a separate regulatory category Normally, regulatory pressure is supposed to reduce demand With privacy, it can also remind people why the product exists Zcash is changing in another way too For years it had one clear job: private payments Now Zcash Shielded Assets are opening the door to private tokens, collectibles and NFT experiments. ZecBit is working on shielded NFT infrastructure while other projects are experimenting with private ownership and sealed-bid auctions That ecosystem is still early, I wouldn’t pretend otherwise But for the first time Zcash is developing a culture layer on top of its privacy layer That gives people another reason to hold and use the network beyond sending money The timing of this rally is probably what makes it stand out most The Fed raised rates The CLARITY Act failed to advance Bitcoin briefly traded around $75K ZEC still climbed roughly 35% While everyone else was waiting for direction from the Senate and the Fed a network built around escaping financial surveillance barely cared Maybe that’s the point For me this move stands on two legs The first is repaired trust after Ironwood The second is growing utility around shielded assets and applications The metric I’ll keep watching isn’t a politician a headline or a mystery whale It’s the amount of ZEC entering and remaining inside shielded pools If that number keeps rising the privacy premium has real usage behind it If it stalls the rally becomes much easier to describe as speculation There is a fair bearish argument here too F2Pool co-founder Chun Wang has described the move as a narrative-driven short squeeze. He has a point privacy still isn’t the default setting and a near-vertical move always attracts leverage and momentum traders The chart is stretched so I’m not treating every green candle as a safe entry $1,400 is the first short-term pivot Below that, $1,250-$1,300 is the main demand area I want to see defended A clean break and hold above the recent high around $1,530 would open $1,650 first followed by a possible extension toward $1,800 Lose $1,100 and the structure starts looking much weaker September 30 is the next date on my calendar, features that aren’t ready for the coming NU7 upgrade by that deadline may be dropped from this release So yes, part of this move may be a squeeze But it’s difficult to call the entire rally empty when the network has repaired its biggest credibility problem shielded usage is growing and a new application layer is beginning to appear at the same time Zcash spent years selling one idea Privacy when you need it The market may finally be deciding that it needs it Not financial advice, just how I’m reading the data and the chart

Arjantit

35,092 просмотров • 7 дней назад