Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

one model built ten webpages in 160 seconds. the next one took 623. same prompt to three models at once: ten standalone HTML pages, one per footballer, shared template and CSS, each self-contained. Ling-3.0-flash: 160s Step 3.7 Flash: 319s MiniMax M2.7: 623s first reasoning token came back in 84ms.... show more

Hanako

12,994 subscribers

34,332 views • 8 days ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

🧵 1/7 I gave a Flash model a written spec for a 3D physics game. It broke the task into steps, wrote the code, used the tools, ran the project, and eventually produced a billiards game I could actually play. Ant Ling Ling-3.0-flash has 124B total parameters, but activates just 5.1B per token. That low activation count didn’t limit it to a simple one-shot demo. It kept a multi-stage 3D build moving from spec to playable result. This test made one thing clear: stop using your biggest model for every step of an agent workflow. The project may be complex. Most of the execution steps inside it don’t need a heavyweight planner to rethink everything from scratch. Ling-3.0-flash is built to execute. Here’s the full hands-on video: 👇

🧵 1/7 I gave a Flash model a written spec for a 3D physics game. It broke the task into steps, wrote the code, used the tools, ran the project, and eventually produced a billiards game I could actually play. Ant Ling Ling-3.0-flash has 124B total parameters, but activates just 5.1B per token. That low activation count didn’t limit it to a simple one-shot demo. It kept a multi-stage 3D build moving from spec to playable result. This test made one thing clear: stop using your biggest model for every step of an agent workflow. The project may be complex. Most of the execution steps inside it don’t need a heavyweight planner to rethink everything from scratch. Ling-3.0-flash is built to execute. Here’s the full hands-on video: 👇

Lonely

119,450 views • 13 days ago

“Next-token prediction” just got a serious rival 🤯 Ant Group just dropped LLaDA 2.1, and it challenges the dominant paradigm of LLMs. Unlike most models that generate one token at a time, LLaDA 2.1 uses diffusion to generate blocks of text in parallel. Why this changes everything: → Global Planning: It effectively sees the "future" while writing the "past" → Parallel Generation: It generates chunks in parallel, not sequentially, being much faster → Massive Efficiency: 16B MoE architecture that only uses ~1.4B active parameters per step. 100% Open Source.

“Next-token prediction” just got a serious rival 🤯 Ant Group just dropped LLaDA 2.1, and it challenges the dominant paradigm of LLMs. Unlike most models that generate one token at a time, LLaDA 2.1 uses diffusion to generate blocks of text in parallel. Why this changes everything: → Global Planning: It effectively sees the "future" while writing the "past" → Parallel Generation: It generates chunks in parallel, not sequentially, being much faster → Massive Efficiency: 16B MoE architecture that only uses ~1.4B active parameters per step. 100% Open Source.

Alvaro Cintas

31,277 views • 5 months ago

I plugged MiniMax (official) M3 into Claude Code in under 5 minutes and gave it a real task. most models need hand-holding every few steps. M3 ran 147 iterations on its own. what it actually did: → took a broken CUDA kernel → ran it for 24 hours, autonomously → pushed hardware efficiency from 7.6% to 71.3% → shipped a 9.4x speedup I didn't touch it once across that whole window. the rest of the spec sheet: → 1M token context window → native multimodal from day one → fully open weights pricing: → $20/month for 1.7B tokens → $50/month for 5.1B tokens → text, image, speech, and music all share one pool usually with this kind of capability the bill catches up fast. M3 keeps both lines low at the same time. that's the part that surprised me. and if this is what M3 does at launch, the full rollout is going to be one to watch. what would you build with 1M context? #MiniMax #MiniMaxM3 #ClaudeCode #LLM

I plugged MiniMax (official) M3 into Claude Code in under 5 minutes and gave it a real task. most models need hand-holding every few steps. M3 ran 147 iterations on its own. what it actually did: → took a broken CUDA kernel → ran it for 24 hours, autonomously → pushed hardware efficiency from 7.6% to 71.3% → shipped a 9.4x speedup I didn't touch it once across that whole window. the rest of the spec sheet: → 1M token context window → native multimodal from day one → fully open weights pricing: → $20/month for 1.7B tokens → $50/month for 5.1B tokens → text, image, speech, and music all share one pool usually with this kind of capability the bill catches up fast. M3 keeps both lines low at the same time. that's the part that surprised me. and if this is what M3 does at launch, the full rollout is going to be one to watch. what would you build with 1M context? #MiniMax #MiniMaxM3 #ClaudeCode #LLM

Nelly;

17,576 views • 2 months ago

first impressions of qwen 3.5 27B dense on a single RTX 3090. 35 tok/s. from 4K all the way to 300K+ context. no speed drop. hermes 4.3 started at 35 and degraded to 15 as context filled. qwen dense holds. MoE held 112 flat. 3x faster but only 3B of 35B active per token. architecture tradeoff. Q4_K_M on 16.7GB. native context 262K. pushed past training limit to 376K before VRAM ceiling on 24GB. tried q8 KV cache at 262K, speed collapsed to 11 tok/s. q4_0 KV is the sweet spot. flash attention mandatory. built in reasoning mode. the model thinks step by step before it answers. full chain of thought surviving Q4 quant. 1,799+ token thinking chains with self correction loops. on a single consumer GPU. gave it one prompt: "build a realtime particle galaxy simulation in one HTML file." 3,340 tokens. 95 seconds. one shot. ran on first load. full reasoning and coding in the video below. optimal config if you want to skip the hours of testing: llama-server -ngl 99 -c 262144 -fa on --cache-type-k q4_0 --cache-type-v q4_0 this is just the warmup. octopus invaders is next: 10 files, 3,400+ lines, zero steering. the prompt hermes quit at 22%. already more impressed than expected. full results coming soon.

first impressions of qwen 3.5 27B dense on a single RTX 3090. 35 tok/s. from 4K all the way to 300K+ context. no speed drop. hermes 4.3 started at 35 and degraded to 15 as context filled. qwen dense holds. MoE held 112 flat. 3x faster but only 3B of 35B active per token. architecture tradeoff. Q4_K_M on 16.7GB. native context 262K. pushed past training limit to 376K before VRAM ceiling on 24GB. tried q8 KV cache at 262K, speed collapsed to 11 tok/s. q4_0 KV is the sweet spot. flash attention mandatory. built in reasoning mode. the model thinks step by step before it answers. full chain of thought surviving Q4 quant. 1,799+ token thinking chains with self correction loops. on a single consumer GPU. gave it one prompt: "build a realtime particle galaxy simulation in one HTML file." 3,340 tokens. 95 seconds. one shot. ran on first load. full reasoning and coding in the video below. optimal config if you want to skip the hours of testing: llama-server -ngl 99 -c 262144 -fa on --cache-type-k q4_0 --cache-type-v q4_0 this is just the warmup. octopus invaders is next: 10 files, 3,400+ lines, zero steering. the prompt hermes quit at 22%. already more impressed than expected. full results coming soon.

Sudo su

120,788 views • 5 months ago

> Opus 4.8 shipped a dial most people won't touch > type /effort in claude code > a menu of compute levels opens up > low for quick stuff, max for the hardest > the one to know: ultracode > xhigh thinking + automatic workflow orchestration > point it at a big task, it plans and runs it end-to-end > same per-token rate at every level > you're not paying more you're choosing the depth > the upgrade isn't the model. it's the dial

> Opus 4.8 shipped a dial most people won't touch > type /effort in claude code > a menu of compute levels opens up > low for quick stuff, max for the hardest > the one to know: ultracode > xhigh thinking + automatic workflow orchestration > point it at a big task, it plans and runs it end-to-end > same per-token rate at every level > you're not paying more you're choosing the depth > the upgrade isn't the model. it's the dial

RetroChainer

18,233 views • 2 months ago

THIS GUY PUT FABLE, GPT 5.6 SOL, AND OPUS IN ONE SHARED CHAT AND HAS THEM TAG EACH OTHER LIKE COWORKERS instead of running three separate models in three separate windows and copy pasting between them, they all sit in one room and talk to each other > you tag whichever model you want and only that one responds > they tag each other too, so fable can hand something to opus, opus can pull in sol for a second opinion > each agent only sees the messages theyre mentioned in, but you get to read everything > he had them running and orchestrating together for more than a day straight the best part is how different each one actually behaves: > fable is the creative product manager. huge context window so it remembers everything, and its the one finding the out of the box solutions > sol is the obsessive tech lead. it will trace every bug to the exact source and stress test every edge case before writing a single line > opus is the workhorse engineer. smart enough for almost anything, and it constantly questions things, which forces the other two to re-examine their own solutions its like managing a team. one plans, one writes the code, one reviews it, and they hand work back and forth without you playing middleman everyone else is arguing about which model is best right now, but the real play is making them work together at the same time

THIS GUY PUT FABLE, GPT 5.6 SOL, AND OPUS IN ONE SHARED CHAT AND HAS THEM TAG EACH OTHER LIKE COWORKERS instead of running three separate models in three separate windows and copy pasting between them, they all sit in one room and talk to each other > you tag whichever model you want and only that one responds > they tag each other too, so fable can hand something to opus, opus can pull in sol for a second opinion > each agent only sees the messages theyre mentioned in, but you get to read everything > he had them running and orchestrating together for more than a day straight the best part is how different each one actually behaves: > fable is the creative product manager. huge context window so it remembers everything, and its the one finding the out of the box solutions > sol is the obsessive tech lead. it will trace every bug to the exact source and stress test every edge case before writing a single line > opus is the workhorse engineer. smart enough for almost anything, and it constantly questions things, which forces the other two to re-examine their own solutions its like managing a team. one plans, one writes the code, one reviews it, and they hand work back and forth without you playing middleman everyone else is arguing about which model is best right now, but the real play is making them work together at the same time

Om Patel

22,969 views • 19 days ago

"Deeply powerful. But also deeply primitive." The AI conversation has split into two camps that are both wrong. Camp one: AI is magic. Six months from AGI. Build everything on top of it. The model will figure it out. Camp two: It's a parlor trick. It hallucinates. It can't reason. The whole thing collapses inside a year. Both miss what's actually happening. Modern AI is genuinely powerful AND genuinely primitive. The same system that drafts a contract in 20 seconds can't tell you, with certainty, what its own confidence is on a single sentence. That's the nature of next-token prediction at scale. The companies that win the next decade hold both ideas at once. They go faster. They also build the rigging that lets them trust what they ship. The ones pretending they already know what this moment is are the ones who'll spend 2027 explaining to the board why a launch they bet the year on shipped wrong. Chamath at Stanford. 43 seconds ↓

"Deeply powerful. But also deeply primitive." The AI conversation has split into two camps that are both wrong. Camp one: AI is magic. Six months from AGI. Build everything on top of it. The model will figure it out. Camp two: It's a parlor trick. It hallucinates. It can't reason. The whole thing collapses inside a year. Both miss what's actually happening. Modern AI is genuinely powerful AND genuinely primitive. The same system that drafts a contract in 20 seconds can't tell you, with certainty, what its own confidence is on a single sentence. That's the nature of next-token prediction at scale. The companies that win the next decade hold both ideas at once. They go faster. They also build the rigging that lets them trust what they ship. The ones pretending they already know what this moment is are the ones who'll spend 2027 explaining to the board why a launch they bet the year on shipped wrong. Chamath at Stanford. 43 seconds ↓

8090

118,067 views • 3 months ago

I'll bite. Let's watch #Falcons Jalon Walker and #49ers Mykel Williams vs #Saints Kelvin Banks Jr. Ewers stepped up into Walker on one play... took over 5 seconds to make a play in one play... and majority of the other pressures we're forced by the right side of the OL. He took on both Mykel and Walker in one play, sack came from the right side. There are multiple instances where Walker and Williams are rushing towards Banks, not once did he let up a sack to that. There is only one bad play in this whole tape, and it is against Mykel Williams... please.

I'll bite. Let's watch #Falcons Jalon Walker and #49ers Mykel Williams vs #Saints Kelvin Banks Jr. Ewers stepped up into Walker on one play... took over 5 seconds to make a play in one play... and majority of the other pressures we're forced by the right side of the OL. He took on both Mykel and Walker in one play, sack came from the right side. There are multiple instances where Walker and Williams are rushing towards Banks, not once did he let up a sack to that. There is only one bad play in this whole tape, and it is against Mykel Williams... please.

Valente ⚜️

67,859 views • 1 year ago

This Chinese developer linked two $2,999 NVIDIA DGX Sparks into one box and runs the full Qwen3-235B at home, after dropping his $1,999-a-month cloud bill to zero. He wired 2 small boxes into a single computer, split a giant 235-billion-parameter model in half between them, and serves it across his own network at about 10 tokens a second, with no internet, no cloud, right there on the desk. No data center, no thousand-dollar graphics cards, no monthly cloud bill. Just him, 2 gold boxes the size of a sandwich, one cable between them, and 1 power strip. And here is the whole payoff. He used to pay the cloud $1,999 a month for the same model, and the meter ticked on every request. Now he paid $5,998 once for 2 boxes, they covered their cost in 3 months, and after that he sends as many requests as he wants for free, only electricity. The two Sparks talk over one fast cable, each holds 128GB of memory, and together they carry the whole model, about 73GB loaded per box, with the chip inside pinned near the limit at 96%. Both boxes work as one and keep trading data over the cable, with no cloud in the loop and no single word leaking out. The ready model sits on one local address, and any app on his network calls it as easily as ChatGPT. And here is how he described, in plain words, what this pair of boxes does: "this is a pair of boxes that holds the huge Qwen3-235B model and serves it to one network. the model is split in half, and each box owns its half. parts: // Box 1 (holds the first half of the model and starts the answer fast, the first word appears in under a second) // Box 2 (holds the second half and writes out the rest, about 10 tokens a second) // Cable (connects the 2 boxes and moves data between them on every step, with no lag) // Address (one local address where any app sends its request, like to a cloud model) // Test (a script that runs big prompts through and measures speed and delays) // Monitor (checks temperature, power draw, and load on both boxes every 2 seconds). the model never goes to the cloud. he only steps in when a box runs hotter than 80 degrees or the cable between them starts dropping data." So the system knows exactly what it is, what it is for, and where its limits are. It knows it has to hold the whole huge model across 2 boxes on its own. It knows it has to answer every request locally, with no meter, no limits, and no internet. It knows the human is only needed when a box overheats or the link between them stalls. → The setup runs around the clock on 2 boxes, each pulling under 60 watts → However many requests he sends, the monthly bill is $0, only electricity → The first box starts the answer in under a second → The second writes text at about 10 tokens a second → One request at a time: 838 tokens in 85 seconds, first word in 0.8s → Two requests at once: 697 tokens in 108 seconds, first word in 0.7s → Both boxes sit at 96% load and warm up to 76-78 degrees And only when a chip in a box runs hotter than 80 degrees or the cable between the 2 Sparks drops data does the system call the owner. And when he himself is out on a run or in a coffee shop, he still reaches his own model at home from his phone: sends a big prompt to the local Qwen3-235B, gets the full answer back in under a minute and a half, with no token meter ticking and no limit to hit. Here is what the test shows on his screen during one of the night runs: "one request at a time: 838 tokens in 84.9 seconds, first word in 0.8s, then 0.1s per token." "two requests at once: 697 tokens in 107.6 seconds, first word in 0.7s, then 0.15s per token." "Box 1: chip at 96% load, 76 degrees, 56 watts, 73GB used in memory." "Box 2: chip at 96% load, 78 degrees, 56 watts, the Qwen3-235B model fully loaded." And while everyone around is paying for AI by the month and bumping into limits, his top-tier model just sits on the desk and works as much as he wants: his own little power plant instead of a forever meter. He has no server rack of his own and no cloud account behind it. Just 2 DGX Spark boxes on a desk, one model split in half between them, one local address, and a folder of prompts next to it. Out of everything I have seen this year, this is the cleanest way to stop paying for AI: $5,998 of hardware on the desk once, $0 a month to the cloud, unlimited forever, and between them 2 gold boxes, 1 cable, and the full Qwen3-235B answering at home with no internet.

This Chinese developer linked two $2,999 NVIDIA DGX Sparks into one box and runs the full Qwen3-235B at home, after dropping his $1,999-a-month cloud bill to zero. He wired 2 small boxes into a single computer, split a giant 235-billion-parameter model in half between them, and serves it across his own network at about 10 tokens a second, with no internet, no cloud, right there on the desk. No data center, no thousand-dollar graphics cards, no monthly cloud bill. Just him, 2 gold boxes the size of a sandwich, one cable between them, and 1 power strip. And here is the whole payoff. He used to pay the cloud $1,999 a month for the same model, and the meter ticked on every request. Now he paid $5,998 once for 2 boxes, they covered their cost in 3 months, and after that he sends as many requests as he wants for free, only electricity. The two Sparks talk over one fast cable, each holds 128GB of memory, and together they carry the whole model, about 73GB loaded per box, with the chip inside pinned near the limit at 96%. Both boxes work as one and keep trading data over the cable, with no cloud in the loop and no single word leaking out. The ready model sits on one local address, and any app on his network calls it as easily as ChatGPT. And here is how he described, in plain words, what this pair of boxes does: "this is a pair of boxes that holds the huge Qwen3-235B model and serves it to one network. the model is split in half, and each box owns its half. parts: // Box 1 (holds the first half of the model and starts the answer fast, the first word appears in under a second) // Box 2 (holds the second half and writes out the rest, about 10 tokens a second) // Cable (connects the 2 boxes and moves data between them on every step, with no lag) // Address (one local address where any app sends its request, like to a cloud model) // Test (a script that runs big prompts through and measures speed and delays) // Monitor (checks temperature, power draw, and load on both boxes every 2 seconds). the model never goes to the cloud. he only steps in when a box runs hotter than 80 degrees or the cable between them starts dropping data." So the system knows exactly what it is, what it is for, and where its limits are. It knows it has to hold the whole huge model across 2 boxes on its own. It knows it has to answer every request locally, with no meter, no limits, and no internet. It knows the human is only needed when a box overheats or the link between them stalls. → The setup runs around the clock on 2 boxes, each pulling under 60 watts → However many requests he sends, the monthly bill is $0, only electricity → The first box starts the answer in under a second → The second writes text at about 10 tokens a second → One request at a time: 838 tokens in 85 seconds, first word in 0.8s → Two requests at once: 697 tokens in 108 seconds, first word in 0.7s → Both boxes sit at 96% load and warm up to 76-78 degrees And only when a chip in a box runs hotter than 80 degrees or the cable between the 2 Sparks drops data does the system call the owner. And when he himself is out on a run or in a coffee shop, he still reaches his own model at home from his phone: sends a big prompt to the local Qwen3-235B, gets the full answer back in under a minute and a half, with no token meter ticking and no limit to hit. Here is what the test shows on his screen during one of the night runs: "one request at a time: 838 tokens in 84.9 seconds, first word in 0.8s, then 0.1s per token." "two requests at once: 697 tokens in 107.6 seconds, first word in 0.7s, then 0.15s per token." "Box 1: chip at 96% load, 76 degrees, 56 watts, 73GB used in memory." "Box 2: chip at 96% load, 78 degrees, 56 watts, the Qwen3-235B model fully loaded." And while everyone around is paying for AI by the month and bumping into limits, his top-tier model just sits on the desk and works as much as he wants: his own little power plant instead of a forever meter. He has no server rack of his own and no cloud account behind it. Just 2 DGX Spark boxes on a desk, one model split in half between them, one local address, and a folder of prompts next to it. Out of everything I have seen this year, this is the cleanest way to stop paying for AI: $5,998 of hardware on the desk once, $0 a month to the cloud, unlimited forever, and between them 2 gold boxes, 1 cable, and the full Qwen3-235B answering at home with no internet.

Blaze

93,871 views • 2 months ago

Most AI research demos show you a polished answer. This one showed me the disagreement that happened before the answer. I gave Ling-3.0-flash Ant Ling a deliberately difficult question: Do four-day workweeks actually increase productivity, or do they simply compress the same workload into fewer days? Instead of asking for a quick summary, I asked it to coordinate five specialist roles: a scientist, a data analyst, a cross-validator, an archivist, and a research writer. Each role had a separate responsibility. The scientist defined the competing hypotheses. The analyst extracted comparable findings. The archivist tracked the sources. The writer could only use approved claims. And the cross-validator had one job: challenge anything that sounded more confident than the evidence allowed. That last role changed the result. The team reviewed 12 sources and challenged six major claims. Three claims were narrowed. One was rejected entirely. Even a widely repeated claim about a 40% productivity increase did not survive the evidence check. That is the part I wanted to see from an AI research workflow. Not just more information, but visible resistance to weak evidence. The final output included: - a direct executive answer - a structured research paper - a source and evidence table - a disagreement log - a six-slide executive deck - a quality-control summary The conclusion was also more useful than a simple yes or no: reduced working hours may maintain productivity and improve wellbeing under certain conditions, while compressing the same workload into fewer days can increase fatigue and intensity. The evidence did not support a universal productivity claim. What impressed me was not that Ling-3.0-flash generated a long response. Plenty of models can do that. It was the way the model maintained multiple roles, evidence standards, objections, citations, and deliverables across one extended workflow, while preserving uncertainty instead of smoothing it away. That makes Ling-3.0-flash especially interesting for work where execution matters as much as reasoning: research, search, coding, document processing, tool use, repeated checks, and other multi-step agent workflows. The strongest AI systems will not use the largest model for every task. They will combine deep planning with fast, cost-efficient execution. Ling-3.0-flash is built for that execution layer. Ling-3.0-flash is now available on OpenRouter and free to use through August 3, 2026. Try it in your coding, search, research, and tool-use workflows. Then show us what you build. Try Ling-3.0-flash: Documentation:

Most AI research demos show you a polished answer. This one showed me the disagreement that happened before the answer. I gave Ling-3.0-flash Ant Ling a deliberately difficult question: Do four-day workweeks actually increase productivity, or do they simply compress the same workload into fewer days? Instead of asking for a quick summary, I asked it to coordinate five specialist roles: a scientist, a data analyst, a cross-validator, an archivist, and a research writer. Each role had a separate responsibility. The scientist defined the competing hypotheses. The analyst extracted comparable findings. The archivist tracked the sources. The writer could only use approved claims. And the cross-validator had one job: challenge anything that sounded more confident than the evidence allowed. That last role changed the result. The team reviewed 12 sources and challenged six major claims. Three claims were narrowed. One was rejected entirely. Even a widely repeated claim about a 40% productivity increase did not survive the evidence check. That is the part I wanted to see from an AI research workflow. Not just more information, but visible resistance to weak evidence. The final output included: - a direct executive answer - a structured research paper - a source and evidence table - a disagreement log - a six-slide executive deck - a quality-control summary The conclusion was also more useful than a simple yes or no: reduced working hours may maintain productivity and improve wellbeing under certain conditions, while compressing the same workload into fewer days can increase fatigue and intensity. The evidence did not support a universal productivity claim. What impressed me was not that Ling-3.0-flash generated a long response. Plenty of models can do that. It was the way the model maintained multiple roles, evidence standards, objections, citations, and deliverables across one extended workflow, while preserving uncertainty instead of smoothing it away. That makes Ling-3.0-flash especially interesting for work where execution matters as much as reasoning: research, search, coding, document processing, tool use, repeated checks, and other multi-step agent workflows. The strongest AI systems will not use the largest model for every task. They will combine deep planning with fast, cost-efficient execution. Ling-3.0-flash is built for that execution layer. Ling-3.0-flash is now available on OpenRouter and free to use through August 3, 2026. Try it in your coding, search, research, and tool-use workflows. Then show us what you build. Try Ling-3.0-flash: Documentation:

Rachel🥥

73,264 views • 13 days ago

Watch King reveals that a random first-time customer bought a $1,700,000 one-of-one AP watch “Random client came in for the first time, nobody, not a referral, at least as far as we know. Never interacted with him, and they asked us for a one-of-one AP, a truly unique piece, and it’s $1.7 million for the watch” “Essentially, I had my team on it, and within three days we found where in the world the watch was. We flew somebody there, we inspected it, we brought it back, we had international armed security, which is a whole thing, and we delivered the watch to the guy at his house, in under a week”

Watch King reveals that a random first-time customer bought a $1,700,000 one-of-one AP watch “Random client came in for the first time, nobody, not a referral, at least as far as we know. Never interacted with him, and they asked us for a one-of-one AP, a truly unique piece, and it’s $1.7 million for the watch” “Essentially, I had my team on it, and within three days we found where in the world the watch was. We flew somebody there, we inspected it, we brought it back, we had international armed security, which is a whole thing, and we delivered the watch to the guy at his house, in under a week”

Mikli

220,865 views • 2 months ago

CLAUDE MADE ME 3 TRADING BOTS IN 15 MINUTES +$2,503 in my wallet the next day, I quit my job that same day. I wrote one prompt and answered a few questions. Claude took three strategies - MACD, RSI + VWAP, CVD divergence. Assigned each bot its own. The first one catches momentum - sees when volume picks up and gets in before the crowd wakes up. The second one trades reversals - waits for everyone to panic and bets against them. The third one scans divergences - when price says one thing but money does another, it follows the money. Built the structure itself, wrote the backtest, ran each strategy on historical data, set up the risk manager, deployed - all on its own. Three bots, three accounts, each trades differently - they don't know about each other. Started with $1, then $5, then $10, then $50, then $2,503 in a day. Citadel, Jane Street, Two Sigma have been trading with bots for years - they don't feel fear, greed, or FOMO - they only listen to algorithms. The market rewards systems.

CLAUDE MADE ME 3 TRADING BOTS IN 15 MINUTES +$2,503 in my wallet the next day, I quit my job that same day. I wrote one prompt and answered a few questions. Claude took three strategies - MACD, RSI + VWAP, CVD divergence. Assigned each bot its own. The first one catches momentum - sees when volume picks up and gets in before the crowd wakes up. The second one trades reversals - waits for everyone to panic and bets against them. The third one scans divergences - when price says one thing but money does another, it follows the money. Built the structure itself, wrote the backtest, ran each strategy on historical data, set up the risk manager, deployed - all on its own. Three bots, three accounts, each trades differently - they don't know about each other. Started with $1, then $5, then $10, then $50, then $2,503 in a day. Citadel, Jane Street, Two Sigma have been trading with bots for years - they don't feel fear, greed, or FOMO - they only listen to algorithms. The market rewards systems.

zostaff

1,947,174 views • 4 months ago

Karpathy just proved the structure beats the model - mapped the whole path from loops to graphs into a 15-page PDF: he pointed one agent at code he'd already hand-optimized. 700 experiments in two days, then he said the single loop was never the target here's the full system: step 1 → write the scorer before anything else. one number, automatic, under 10 minutes. no scorer, no loop step 2 → split the files. one the agent edits, one holding the scorer it can never reach step 3 → fix the wall clock per attempt. comparability comes from the budget, not the attempts step 4 → write the never-stop line verbatim or your agent becomes a slower human step 5 → add a second agent only after the first ships real commits. one git worktree each or they overwrite each other step 6 → at thirty agents, isolation costs more than it saves. that's when they need one shared DAG instead of thirty private histories 2.02 hours to 1.80 hours on code its author had already tuned. 90,900 stars in four months. read this PDF and paste it into your Claude, you won't regret it bookmark, then read the article on building graphs from scratch ↓

Karpathy just proved the structure beats the model - mapped the whole path from loops to graphs into a 15-page PDF: he pointed one agent at code he'd already hand-optimized. 700 experiments in two days, then he said the single loop was never the target here's the full system: step 1 → write the scorer before anything else. one number, automatic, under 10 minutes. no scorer, no loop step 2 → split the files. one the agent edits, one holding the scorer it can never reach step 3 → fix the wall clock per attempt. comparability comes from the budget, not the attempts step 4 → write the never-stop line verbatim or your agent becomes a slower human step 5 → add a second agent only after the first ships real commits. one git worktree each or they overwrite each other step 6 → at thirty agents, isolation costs more than it saves. that's when they need one shared DAG instead of thirty private histories 2.02 hours to 1.80 hours on code its author had already tuned. 90,900 stars in four months. read this PDF and paste it into your Claude, you won't regret it bookmark, then read the article on building graphs from scratch ↓

Carnage

53,533 views • 11 days ago

HERMES AGENT CROSSED 140,000 GITHUB STARS IN 3 MONTHS AND JUST BECAME THE MOST USED AGENT IN THE WORLD. Most AI agents forget everything between sessions. Hermes writes its own skills from experience. Next time it runs the skill, improves it, and gets faster. Independent benchmarks show agents with 20+ self-created skills complete similar tasks 40% faster than fresh instances. Qwen 3.6 where the 35B version outperforms last year's 120B models at one third the memory footprint. DGX Spark with 128GB unified memory running everything locally at $0 per month after hardware. The setup takes 30 minutes. LM Studio plus Qwen 3.6 27B for the model server. One install script for Hermes. One config connecting them. Set context window to 65,536 tokens or nothing works. After one month of daily use your skills directory has 20 to 50 learned workflows. Your Hermes is genuinely different from anyone else's.

HERMES AGENT CROSSED 140,000 GITHUB STARS IN 3 MONTHS AND JUST BECAME THE MOST USED AGENT IN THE WORLD. Most AI agents forget everything between sessions. Hermes writes its own skills from experience. Next time it runs the skill, improves it, and gets faster. Independent benchmarks show agents with 20+ self-created skills complete similar tasks 40% faster than fresh instances. Qwen 3.6 where the 35B version outperforms last year's 120B models at one third the memory footprint. DGX Spark with 128GB unified memory running everything locally at $0 per month after hardware. The setup takes 30 minutes. LM Studio plus Qwen 3.6 27B for the model server. One install script for Hermes. One config connecting them. Set context window to 65,536 tokens or nothing works. After one month of daily use your skills directory has 20 to 50 learned workflows. Your Hermes is genuinely different from anyone else's.

Lummox

69,676 views • 2 months ago

With Kimi K3 Day-0 on vLLM: Open Frontier Intelligence for Everyone 🚀 At 2.8 trillion parameters, Moonshot AI's Kimi K3 is one of the most powerful open-weight models ever released. Starting today, you can serve it on vLLM the moment the weights are public. What K3 brings: 🧠 2.8T-parameter Mixture-of-Experts (16 of 896 experts active per token) 📚 1M-token context window 🛠️ Native multimodal understanding, including vision ⚡ Kimi Delta Attention: a hybrid of linear and full attention that makes million-token context affordable Huge thank you to Kimi.ai AI for the model release and partnership, Inferact for leading the vLLM optimizations, and to our partners at NVIDIA, AMD, and the broader vLLM community. 1/6

With Kimi K3 Day-0 on vLLM: Open Frontier Intelligence for Everyone 🚀 At 2.8 trillion parameters, Moonshot AI's Kimi K3 is one of the most powerful open-weight models ever released. Starting today, you can serve it on vLLM the moment the weights are public. What K3 brings: 🧠 2.8T-parameter Mixture-of-Experts (16 of 896 experts active per token) 📚 1M-token context window 🛠️ Native multimodal understanding, including vision ⚡ Kimi Delta Attention: a hybrid of linear and full attention that makes million-token context affordable Huge thank you to Kimi.ai AI for the model release and partnership, Inferact for leading the vLLM optimizations, and to our partners at NVIDIA, AMD, and the broader vLLM community. 1/6

vLLM

101,621 views • 12 days ago

Two data points dropped in the last few months that should terrify every software company that thinks its codebase is a moat. First, one engineer at Cloudflare, working with Claude via AI agents, rebuilt 94% of Next.js, one of the most widely used frontend frameworks on the internet, built over 10 years by a large engineering team in a single week. Total cost was $1,100 in API tokens. The result, called Vinext, is a drop-in replacement that builds production apps up to 4x faster and produces client bundles 57% smaller and customers are already running it in production. Second is Cursor CEO Michael Truell deployed a swarm of hundreds of GPT-5.2 agents that ran uninterrupted for an entire week and built a fully functional web browser from scratch called FastRender. 3 million lines of code, thousands of files and a custom Rust rendering engine with HTML parsing, CSS layout, text shaping, and a custom JavaScript VM. Total cost was roughly $30,000. For context, Google has spent billions of dollars and decades of engineering building Chrome. And the benchmarks say by next year, you will be able to one-shot prompt anything. The moat that software companies spent decades building, the complexity of their codebase, the years it would take a competitor to replicate it, the switching costs that moat assumed humans were the unit of production. AI does not care how long it took you to build it, it only cares how long it takes to rebuild it. And right now, the answer is one week.

Two data points dropped in the last few months that should terrify every software company that thinks its codebase is a moat. First, one engineer at Cloudflare, working with Claude via AI agents, rebuilt 94% of Next.js, one of the most widely used frontend frameworks on the internet, built over 10 years by a large engineering team in a single week. Total cost was $1,100 in API tokens. The result, called Vinext, is a drop-in replacement that builds production apps up to 4x faster and produces client bundles 57% smaller and customers are already running it in production. Second is Cursor CEO Michael Truell deployed a swarm of hundreds of GPT-5.2 agents that ran uninterrupted for an entire week and built a fully functional web browser from scratch called FastRender. 3 million lines of code, thousands of files and a custom Rust rendering engine with HTML parsing, CSS layout, text shaping, and a custom JavaScript VM. Total cost was roughly $30,000. For context, Google has spent billions of dollars and decades of engineering building Chrome. And the benchmarks say by next year, you will be able to one-shot prompt anything. The moat that software companies spent decades building, the complexity of their codebase, the years it would take a competitor to replicate it, the switching costs that moat assumed humans were the unit of production. AI does not care how long it took you to build it, it only cares how long it takes to rebuild it. And right now, the answer is one week.

Milk Road AI

16,781 views • 3 months ago

$Mark Zuckerberg is explaining one of the most misunderstood dynamics in AI and it has direct investment implications (Save this). The concept he's describing is model distillation, and it's one of the most important techniques to emerge in AI over the past year. Here's how it works. You train a massive, enormously expensive model, in Meta's case, Llama 4 Behemoth, a 2 trillion parameter teacher model and then you use that model to teach a much smaller, cheaper model. The smaller model inherits roughly 90 to 95% of the intelligence of the giant while running at 10% of the cost and on a fraction of the compute. Meta already did this with the Llama 4 family and Behemoth serves as the teacher. Llama 4 Scout and Maverick, the publicly released open-source models were distilled from it. Scout runs on a single H100 GPU with a 10 million token context window and outperforms models that cost far more to operate. Maverick, at 17 billion active parameters, rivals DeepSeek V3 in coding at half the parameter count and beats GPT-4o on multimodal benchmarks. Both are completely free for commercial use. What Zuckerberg is pointing at is a structural shift in how AI gets deployed in the real world. Companies aren't taking a frontier model off the shelf and running it as-is but rather taking open-source models, fine-tuning them on their own proprietary data, distilling them into even smaller custom models tailored to their specific use case, and running them on infrastructure they control at a fraction of the cost of a closed frontier API. The investment implication of this is significant and runs in two directions. For Meta specifically, this is a strategic masterstroke. Every company that builds on Llama, fine-tunes it, distills it, or deploys it through their infrastructure is pulling into Meta's orbit while Meta builds the most powerful open teacher model. The ecosystem of companies using it grows and that ecosystem generates commercial activity across Meta's platforms and data services. Meta's AI research benefits from billions of real world deployment signals and it's a flywheel that closed model providers cannot replicate because their strategy requires charging per token, which is now a 65x cost disadvantage against the open-source alternative. For the broader market, distillation changes the economics of inference in a way that has barely been priced in. As intelligence becomes extractable into smaller and cheaper models, the absolute demand for compute doesn't decline but rather it explodes, because now the number of applications that are economically viable expands by orders of magnitude. Every task that was previously too expensive to automate at $3.25 per call becomes viable at $0.05 that means more total token usage, more total GPU utilization, and more demand for the infrastructure companies, the Nebiuses, the GE Vernovas, the Constellation Energies that supply the underlying compute and power.$

Mark Zuckerberg is explaining one of the most misunderstood dynamics in AI and it has direct investment implications (Save this). The concept he's describing is model distillation, and it's one of the most important techniques to emerge in AI over the past year. Here's how it works. You train a massive, enormously expensive model, in Meta's case, Llama 4 Behemoth, a 2 trillion parameter teacher model and then you use that model to teach a much smaller, cheaper model. The smaller model inherits roughly 90 to 95% of the intelligence of the giant while running at 10% of the cost and on a fraction of the compute. Meta already did this with the Llama 4 family and Behemoth serves as the teacher. Llama 4 Scout and Maverick, the publicly released open-source models were distilled from it. Scout runs on a single H100 GPU with a 10 million token context window and outperforms models that cost far more to operate. Maverick, at 17 billion active parameters, rivals DeepSeek V3 in coding at half the parameter count and beats GPT-4o on multimodal benchmarks. Both are completely free for commercial use. What Zuckerberg is pointing at is a structural shift in how AI gets deployed in the real world. Companies aren't taking a frontier model off the shelf and running it as-is but rather taking open-source models, fine-tuning them on their own proprietary data, distilling them into even smaller custom models tailored to their specific use case, and running them on infrastructure they control at a fraction of the cost of a closed frontier API. The investment implication of this is significant and runs in two directions. For Meta specifically, this is a strategic masterstroke. Every company that builds on Llama, fine-tunes it, distills it, or deploys it through their infrastructure is pulling into Meta's orbit while Meta builds the most powerful open teacher model. The ecosystem of companies using it grows and that ecosystem generates commercial activity across Meta's platforms and data services. Meta's AI research benefits from billions of real world deployment signals and it's a flywheel that closed model providers cannot replicate because their strategy requires charging per token, which is now a 65x cost disadvantage against the open-source alternative. For the broader market, distillation changes the economics of inference in a way that has barely been priced in. As intelligence becomes extractable into smaller and cheaper models, the absolute demand for compute doesn't decline but rather it explodes, because now the number of applications that are economically viable expands by orders of magnitude. Every task that was previously too expensive to automate at $3.25 per call becomes viable at $0.05 that means more total token usage, more total GPU utilization, and more demand for the infrastructure companies, the Nebiuses, the GE Vernovas, the Constellation Energies that supply the underlying compute and power.

Milk Road AI

27,908 views • 1 month ago

Qwen 3.7 Max looks incredible on paper. Then I spent $43 in 15 minutes of vibe coding. Best Chinese model on every benchmark. #1 on SWE-Bench Pro. #4 on BridgeBench UI. The scores are legit. But in a real workflow it came back with 15 errors on a single task. No subsidies like Claude Code or Codex. Cheaper per token, more expensive per task. Chinese models are getting closer. But "closer" still means you're burning more time and more money to get the same result GPT 5.5 one shots. Full review and BridgeBench breakdown below.

Qwen 3.7 Max looks incredible on paper. Then I spent $43 in 15 minutes of vibe coding. Best Chinese model on every benchmark. #1 on SWE-Bench Pro. #4 on BridgeBench UI. The scores are legit. But in a real workflow it came back with 15 errors on a single task. No subsidies like Claude Code or Codex. Cheaper per token, more expensive per task. Chinese models are getting closer. But "closer" still means you're burning more time and more money to get the same result GPT 5.5 one shots. Full review and BridgeBench breakdown below.

BridgeMind

15,246 views • 2 months ago

We generated this scene with just one trial on Seedance Enterprise 2.5. We had run this same shot on other models and it took us around 8 attempts before anything came back usable. Generations are billed one at a time so 8 tries to 1 is the entire cost story. 30 seconds, single shot, no stitching, what used to be a day of rerolling is now a single pass, which means the budget goes into testing new concepts instead of rescuing one. Watch the water as the creature moves through it, and watch the camera fall behind during the escape. The realistic physics, smooth camera motion, and shot consistency all came directly from the model.None of that was added in post. We used Seedream 5.0 for the visual ideation and Seedance 2.5 for the video. Both are running on lumina at 4K, up to 50 references to hold a character or product steady across a shot, and region editing to fix one part of a frame without regenerating the whole thing. Seedance 2.5 API is about to launch on BytePlus, and now you can try it directly on Lumina.

We generated this scene with just one trial on Seedance Enterprise 2.5. We had run this same shot on other models and it took us around 8 attempts before anything came back usable. Generations are billed one at a time so 8 tries to 1 is the entire cost story. 30 seconds, single shot, no stitching, what used to be a day of rerolling is now a single pass, which means the budget goes into testing new concepts instead of rescuing one. Watch the water as the creature moves through it, and watch the camera fall behind during the escape. The realistic physics, smooth camera motion, and shot consistency all came directly from the model.None of that was added in post. We used Seedream 5.0 for the visual ideation and Seedance 2.5 for the video. Both are running on lumina at 4K, up to 50 references to hold a character or product steady across a shot, and region editing to fix one part of a frame without regenerating the whole thing. Seedance 2.5 API is about to launch on BytePlus, and now you can try it directly on Lumina.

Future Stacked

85,011 views • 3 days ago

A 24-year-old launched a faceless YouTube channel 13 days ago. In the first month, he has already accumulated $4,000, but he has not shot a single frame.. His investment: a $20 subscription to Claude. No camera, no editor, no voice of his own. THE SYSTEM WORKS LIKE THIS: - Claude writes the full script from one topic prompt - an AI voice narrates it - an AI tool generates the animation and the visuals - one template, repeated forever: "the economics of owning an airport" / "…a private jet" / "…a hospital" - it publishes on a schedule while he sleeps The first 12 days, from zero: > "Why a Fan Can Kill You" -> 272,000 views > second video -> 136,000 views in 6 days > subscribers -> 4,300 > month-one pace -> ~$4,000 And it's replicable. One of his students launched "Mr. Finance" on the same template 13 days ago: > 10,800 subscribers > finance-niche RPM -> ~$7 per 1,000 views > one video on the economics of shopping malls -> $500+ on its own He didn't become a creator. He built a press that prints episodes. The whole stack is public. Everyone has the same $20 Claude. He just pointed it at a niche and refused to stop publishing.

A 24-year-old launched a faceless YouTube channel 13 days ago. In the first month, he has already accumulated $4,000, but he has not shot a single frame.. His investment: a $20 subscription to Claude. No camera, no editor, no voice of his own. THE SYSTEM WORKS LIKE THIS: - Claude writes the full script from one topic prompt - an AI voice narrates it - an AI tool generates the animation and the visuals - one template, repeated forever: "the economics of owning an airport" / "…a private jet" / "…a hospital" - it publishes on a schedule while he sleeps The first 12 days, from zero: > "Why a Fan Can Kill You" -> 272,000 views > second video -> 136,000 views in 6 days > subscribers -> 4,300 > month-one pace -> ~$4,000 And it's replicable. One of his students launched "Mr. Finance" on the same template 13 days ago: > 10,800 subscribers > finance-niche RPM -> ~$7 per 1,000 views > one video on the economics of shopping malls -> $500+ on its own He didn't become a creator. He built a press that prints episodes. The whole stack is public. Everyone has the same $20 Claude. He just pointed it at a niche and refused to stop publishing.

Ridark

142,163 views • 18 days ago