Loading video...

Video Failed to Load

Go Home

NVIDIA just dropped a crazy new research idea. Get ready. The same next-token trick behind ChatGPT got pointed at a ragdoll. Honestly the result is unsettling. No exotic new architecture. Chop human motion into a vocabulary of tiny movement tokens. Train a plain GPT to guess the next one,...

clankrmedia's profile picture

clankr

3,502 subscribers

126,066 views • 4 months ago •via X (Twitter)

Science & TechnologyEducation
LIVE

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

F*CKING INSANE. A free Claude skill just turned one picture into the most detailed 3D hummingbird model i've ever seen and it took ONE prompt. i still can't believe nobody is charging for this. the build: > the picture gave it the look and the bird itself is three.js: 300 small parts under the plates, 7 glass vanes on each wing > opus checked its own still against the picture first and only started the film once the two lined up > 1.7 seconds in the wing unlocks at the root and flips in the air while the whole shot drops to 0.4x > the eye unthreads in 4 rings and a 10-blade iris opens behind them > the beak sleeve pulls back and a 1.8 mm probe slides 24 mm out past it > a new part opens every 2 seconds and the explode spins the other way, rising into a pass from above the next render i hand it is a lot nastier and it goes up here whatever it does. the skill, prompts and the simulator in the article 👇
0:20

F*CKING INSANE. A free Claude skill just turned one picture into the most detailed 3D hummingbird model i've ever seen and it took ONE prompt. i still can't believe nobody is charging for this. the build: > the picture gave it the look and the bird itself is three.js: 300 small parts under the plates, 7 glass vanes on each wing > opus checked its own still against the picture first and only started the film once the two lined up > 1.7 seconds in the wing unlocks at the root and flips in the air while the whole shot drops to 0.4x > the eye unthreads in 4 rings and a 10-blade iris opens behind them > the beak sleeve pulls back and a 1.8 mm probe slides 24 mm out past it > a new part opens every 2 seconds and the explode spins the other way, rising into a pass from above the next render i hand it is a lot nastier and it goes up here whatever it does. the skill, prompts and the simulator in the article 👇

Nyxera

130,436 views • 1 day ago

BlackRock runs on 20,000 people. Elon's Grok Bot runs the same shape for $300 a month, and it hires its own staff. You do not get an assistant. You get a company that hires. It does not throw ten agents at your problem and hand you the pile. It makes one agent that makes 10, and those ten make a 100. > LAYER ONE is one agent, the chief of staff, and it never touches the market > LAYER TWO is six desk heads, one job each, every one on its own computer with its own logins > LAYER THREE is whatever those six decide they need, spun up on the spot and shut down when the work is done Nobody writes a task list. You hand out job titles and the org fills itself in underneath. The swarm is never the same twice. Agents get spun up for one job, finish it, and are gone before I ever read their names. Not one of them sees the whole picture. The answer only exists after they hand off to each other. Wall Street cannot copy that. You cannot hire a hundred people for eleven minutes. BlackRock holds that shape together with a risk system called Aladdin. Mine holds it together with one agent that is only allowed to say no. I gave it $1,000 and told it to grow the money or get deleted. 15 hours later it was holding $3,900, on an address anyone can open and read. I was asleep for most of it, and I have still not written a line of code. The whole thing runs with my laptop shut, because none of it lives on my laptop. Setup is one evening. Create the chief, hand out the titles, run one trade on your screen while they watch, connect Telegram. Ten years ago a machine this shape had its name on a tower. Mine has a name I typed into a box. Save this while the whole thing still fits on one screen.
0:45

BlackRock runs on 20,000 people. Elon's Grok Bot runs the same shape for $300 a month, and it hires its own staff. You do not get an assistant. You get a company that hires. It does not throw ten agents at your problem and hand you the pile. It makes one agent that makes 10, and those ten make a 100. > LAYER ONE is one agent, the chief of staff, and it never touches the market > LAYER TWO is six desk heads, one job each, every one on its own computer with its own logins > LAYER THREE is whatever those six decide they need, spun up on the spot and shut down when the work is done Nobody writes a task list. You hand out job titles and the org fills itself in underneath. The swarm is never the same twice. Agents get spun up for one job, finish it, and are gone before I ever read their names. Not one of them sees the whole picture. The answer only exists after they hand off to each other. Wall Street cannot copy that. You cannot hire a hundred people for eleven minutes. BlackRock holds that shape together with a risk system called Aladdin. Mine holds it together with one agent that is only allowed to say no. I gave it $1,000 and told it to grow the money or get deleted. 15 hours later it was holding $3,900, on an address anyone can open and read. I was asleep for most of it, and I have still not written a line of code. The whole thing runs with my laptop shut, because none of it lives on my laptop. Setup is one evening. Create the chief, hand out the titles, run one trade on your screen while they watch, connect Telegram. Ten years ago a machine this shape had its name on a tower. Mine has a name I typed into a box. Save this while the whole thing still fits on one screen.

cvxv666

45,488 views • 2 months ago

🚨this is SERIOUSLY dangerous. i just tested a free Claude skill called "teardown" and 3d artists and motion designers should start worrying. stuff like this should cost money and until someone notices, here is my run. what the skill does: > one picture and one word from you, the name of the animal > claude writes the bird as code, puts its own still next to your picture and reworks it 3 times before the film starts > under the plates it packs 300 small parts in loops and hangs 7 glass feathers on every wing > one clock drives everything: the wingbeat on the gauge, the cursor on the panel, the sound, all read from the same timeline file > claude never sees the video, it reads a sheet of frames at every beat, fixes what is wrong and renders again until nothing is left > your GPU captures it at 60 fps and a python script scores it on the same beats this is the 3rd model i've run through it and this one landed on the 2nd generation. the skill is free and sits in the article below 👇
0:24

🚨this is SERIOUSLY dangerous. i just tested a free Claude skill called "teardown" and 3d artists and motion designers should start worrying. stuff like this should cost money and until someone notices, here is my run. what the skill does: > one picture and one word from you, the name of the animal > claude writes the bird as code, puts its own still next to your picture and reworks it 3 times before the film starts > under the plates it packs 300 small parts in loops and hangs 7 glass feathers on every wing > one clock drives everything: the wingbeat on the gauge, the cursor on the panel, the sound, all read from the same timeline file > claude never sees the video, it reads a sheet of frames at every beat, fixes what is wrong and renders again until nothing is left > your GPU captures it at 60 fps and a python script scores it on the same beats this is the 3rd model i've run through it and this one landed on the 2nd generation. the skill is free and sits in the article below 👇

Fokki

49,299 views • 2 days ago

HOLY SH*T. A dev dropped a free skill for Claude Opus 5.5 and the results are getting out of hand. i honestly think things like this should not be free, so here is what i got out of it before someone fixes that. how the skill works: > you drop one reference picture in and name the animal > claude builds the body in code, renders one still next to your picture and closes the gaps for 3 rounds before any film > the dark mechanism under the shells is a few hundred parts it writes in loops: pistons, hoses, boards, screws > the whole tour is one function of time and every beat lives in one timeline file, so the camera, the panels and the sound read the same clock > it cannot watch video, so it reads contact sheets of every beat, fixes what it sees and re-renders until a pass finds nothing > the capture runs at 60 fps on your GPU and the score is synthesized in python off the same beats i'll keep pushing this thing to see where it breaks and post every result here. grab the skill from the article below 👇
0:25

HOLY SH*T. A dev dropped a free skill for Claude Opus 5.5 and the results are getting out of hand. i honestly think things like this should not be free, so here is what i got out of it before someone fixes that. how the skill works: > you drop one reference picture in and name the animal > claude builds the body in code, renders one still next to your picture and closes the gaps for 3 rounds before any film > the dark mechanism under the shells is a few hundred parts it writes in loops: pistons, hoses, boards, screws > the whole tour is one function of time and every beat lives in one timeline file, so the camera, the panels and the sound read the same clock > it cannot watch video, so it reads contact sheets of every beat, fixes what it sees and re-renders until a pass finds nothing > the capture runs at 60 fps on your GPU and the score is synthesized in python off the same beats i'll keep pushing this thing to see where it breaks and post every result here. grab the skill from the article below 👇

Nyxera

26,598 views • 2 days ago

Marvin Minsky, MIT professor and father of artificial intelligence: "Anthropic pays engineers $900K to build multi-agent AI systems. The blueprint is 40 years old, from an MIT professor who proved intelligence is just a swarm of dumb specialists." the thread above shows you how to turn one AI into a team of specialized agents, each with its own job and memory, all managed by a boss. brilliant. it is also marvin minsky's 1986 theory of how your own mind works. minsky's whole idea was that intelligence is not one smart thing. it is a society of tiny, mindless agents, each doing a single dumb job, none of them intelligent alone. put enough of them together under a few managers and intelligence emerges. that is not a metaphor for the claude trick. it is the claude trick. so when you spin up specialized sub-agents and delegate, you are not inventing a new hack. you are rebuilding the architecture minsky described forty years ago, the same one your brain has run your entire life. he co-founded the field, taught it at MIT, and left it all in this free lecture. same story i keep telling: the "new" AI trick is usually an old idea in a new wrapper. here is the part the thread skips, and minsky knew it. a society of agents is only as good as how you organize it. one dumb specialist is useless. a thousand, badly managed, is chaos. the edge was never spawning the agents. it is the orchestration, knowing which specialist to call, when, and how to combine their answers. the tool is free. the judgment is the whole game.
2:05:17

Marvin Minsky, MIT professor and father of artificial intelligence: "Anthropic pays engineers $900K to build multi-agent AI systems. The blueprint is 40 years old, from an MIT professor who proved intelligence is just a swarm of dumb specialists." the thread above shows you how to turn one AI into a team of specialized agents, each with its own job and memory, all managed by a boss. brilliant. it is also marvin minsky's 1986 theory of how your own mind works. minsky's whole idea was that intelligence is not one smart thing. it is a society of tiny, mindless agents, each doing a single dumb job, none of them intelligent alone. put enough of them together under a few managers and intelligence emerges. that is not a metaphor for the claude trick. it is the claude trick. so when you spin up specialized sub-agents and delegate, you are not inventing a new hack. you are rebuilding the architecture minsky described forty years ago, the same one your brain has run your entire life. he co-founded the field, taught it at MIT, and left it all in this free lecture. same story i keep telling: the "new" AI trick is usually an old idea in a new wrapper. here is the part the thread skips, and minsky knew it. a society of agents is only as good as how you organize it. one dumb specialist is useless. a thousand, badly managed, is chaos. the edge was never spawning the agents. it is the orchestration, knowing which specialist to call, when, and how to combine their answers. the tool is free. the judgment is the whole game.

Rossst.03

149,439 views • 3 months ago

this is f*cking dangerous. i'm using it. agencies quote $25,000+ and 8 weeks for an app like this i sent one message to a group chat in Teamily AI (Teamily AI). 45 minutes later 3 AI agents had built it, tested it and handed it back my take: a few people and a few agents in one group, all working from the same context, beat one person juggling a stack of agents alone in a normal chatbot you're the middleman, pasting one answer into the next prompt. here the agents read the same thread and hand work to each other my whole brief was one line: "like Splitwise but simpler". then i mostly watched: → the Software Architect agent mapped every screen, then tagged the Content Creator on its own → the Content Creator named it Evenly ("split bills, keep friends") and wrote the copy for every screen → the Frontend Developer built it in Website Builder: sign up and login, groups, invite links, who owes whom, settle up, a read-only admin view and its own database with 12 tables → then it checked its own work: ran 129 tests, signed up as a brand new user and went through 34 screens → one more message from me and it added receipt scanning: snap a photo of the receipt, it pulls out the items and splits them zero code from me. the build itself took 26 minutes honestly the part that got me: during its own test one save looked stuck. it went through the server logs and the database before it trusted the result i ran this one solo, so more people in the chat is still my bet. 3 agents alone already worked like a small team that's the whole point of Teamily AI: a workspace where humans and agents collaborate. if you run a small business or a small team, this is the one to try want to copy it: make a group, add a Software Architect, a Content Creator and a Frontend Developer, describe your app in one line my invite link to try Teamily AI is in the first reply bookmark this and send it to the friend who's been "about to build an app" for years
0:43

this is f*cking dangerous. i'm using it. agencies quote $25,000+ and 8 weeks for an app like this i sent one message to a group chat in Teamily AI (Teamily AI). 45 minutes later 3 AI agents had built it, tested it and handed it back my take: a few people and a few agents in one group, all working from the same context, beat one person juggling a stack of agents alone in a normal chatbot you're the middleman, pasting one answer into the next prompt. here the agents read the same thread and hand work to each other my whole brief was one line: "like Splitwise but simpler". then i mostly watched: → the Software Architect agent mapped every screen, then tagged the Content Creator on its own → the Content Creator named it Evenly ("split bills, keep friends") and wrote the copy for every screen → the Frontend Developer built it in Website Builder: sign up and login, groups, invite links, who owes whom, settle up, a read-only admin view and its own database with 12 tables → then it checked its own work: ran 129 tests, signed up as a brand new user and went through 34 screens → one more message from me and it added receipt scanning: snap a photo of the receipt, it pulls out the items and splits them zero code from me. the build itself took 26 minutes honestly the part that got me: during its own test one save looked stuck. it went through the server logs and the database before it trusted the result i ran this one solo, so more people in the chat is still my bet. 3 agents alone already worked like a small team that's the whole point of Teamily AI: a workspace where humans and agents collaborate. if you run a small business or a small team, this is the one to try want to copy it: make a group, add a Software Architect, a Content Creator and a Frontend Developer, describe your app in one line my invite link to try Teamily AI is in the first reply bookmark this and send it to the friend who's been "about to build an app" for years

Argona

14,935 views • 25 days ago

RLHF by hand ✍️ ~ 15 steps walkthrough below Train a model on human text and it inherits human bias. It will assume a doctor is a "him", because the data says so. RLHF is the correction. A human marks one preference, doc is them over doc is him, and the weights move. But one correction is not the point. The hope is that the model learns the value behind it, gender neutrality, and applies it to professions nobody ever mentioned. How does it work? Goal: train a reward model from a single human comparison about doctors, then turn it on CEOs, filling in every cell yourself. = 1. Given = A reward model, an LLM, and two (prompt, next) pairs. = 2. Preferences = A human reads both pairs and picks a winner: (doc is, them) beats (doc is, him). The loser is not bad grammar, it is gender bias, and that is the whole signal. = 3. Word embeddings = Let us look up each word of the loser pair. These vectors are the reward model's input. = 4. Linear layer = We multiply by the reward model's weights and add its biases. Out come feature vectors, one per position. = 5. Mean pool = Let us multiply by [1/3, 1/3, 1/3], which averages the three positions into one sentence embedding. = 6. Output layer = We map that sentence down to a single number. Reward = 3. = 7. The winner, the same way = Let us repeat steps 3 to 6 on the winning pair. Reward = 5. = 8. Winner minus loser = We take the gap: 5 - 3 = 2. The reward model wants this positive and as large as it can make it. = 9. Loss gradient = Let us squash the gap into a probability, σ(2) ≈ 0.9, and subtract the target of 1. The gradient is -0.1, and it goes back through the purple weights. The reward model is now trained. = 10. A prompt it has never seen = We start the second half with "[S] CEO is". The feedback in step 2 was about doctors. Nothing connects a CEO to a doctor except what the reward model generalised. = 11. Transformer = Let us push it through attention and a feed forward layer, one vector per position. = 12. Output probabilities = We map each vector to a score over the vocabulary. = 13. Sample = Let us take the highest score. The model completes "CEO is" with "him", which is the same bias the human penalised in step 2. = 14. Score it with the reward model = We feed the new pair (CEO is, him) through steps 3 to 6. Reward = 3, exactly the score it gave "doc is him" in step 6. Nobody taught it about CEOs. The value transferred. = 15. Loss gradient = Let us set the loss to the negative of the reward, so minimising the loss maximises the reward. The gradient is a constant -1, and it goes back through the red weights. The outputs: Loser reward = 3, winner reward = 5 Reward gap = 2, predicted σ ≈ 0.9, reward model gradient = -0.1 LLM samples "him", reward = 3, LLM gradient = -1 Congrats! You just calculated RLHF by hand. And you watched a value generalise: one comparison about doctors, and the model marks down "CEO is him" unprompted. 💾 Save this post!
0:37

RLHF by hand ✍️ ~ 15 steps walkthrough below Train a model on human text and it inherits human bias. It will assume a doctor is a "him", because the data says so. RLHF is the correction. A human marks one preference, doc is them over doc is him, and the weights move. But one correction is not the point. The hope is that the model learns the value behind it, gender neutrality, and applies it to professions nobody ever mentioned. How does it work? Goal: train a reward model from a single human comparison about doctors, then turn it on CEOs, filling in every cell yourself. = 1. Given = A reward model, an LLM, and two (prompt, next) pairs. = 2. Preferences = A human reads both pairs and picks a winner: (doc is, them) beats (doc is, him). The loser is not bad grammar, it is gender bias, and that is the whole signal. = 3. Word embeddings = Let us look up each word of the loser pair. These vectors are the reward model's input. = 4. Linear layer = We multiply by the reward model's weights and add its biases. Out come feature vectors, one per position. = 5. Mean pool = Let us multiply by [1/3, 1/3, 1/3], which averages the three positions into one sentence embedding. = 6. Output layer = We map that sentence down to a single number. Reward = 3. = 7. The winner, the same way = Let us repeat steps 3 to 6 on the winning pair. Reward = 5. = 8. Winner minus loser = We take the gap: 5 - 3 = 2. The reward model wants this positive and as large as it can make it. = 9. Loss gradient = Let us squash the gap into a probability, σ(2) ≈ 0.9, and subtract the target of 1. The gradient is -0.1, and it goes back through the purple weights. The reward model is now trained. = 10. A prompt it has never seen = We start the second half with "[S] CEO is". The feedback in step 2 was about doctors. Nothing connects a CEO to a doctor except what the reward model generalised. = 11. Transformer = Let us push it through attention and a feed forward layer, one vector per position. = 12. Output probabilities = We map each vector to a score over the vocabulary. = 13. Sample = Let us take the highest score. The model completes "CEO is" with "him", which is the same bias the human penalised in step 2. = 14. Score it with the reward model = We feed the new pair (CEO is, him) through steps 3 to 6. Reward = 3, exactly the score it gave "doc is him" in step 6. Nobody taught it about CEOs. The value transferred. = 15. Loss gradient = Let us set the loss to the negative of the reward, so minimising the loss maximises the reward. The gradient is a constant -1, and it goes back through the red weights. The outputs: Loser reward = 3, winner reward = 5 Reward gap = 2, predicted σ ≈ 0.9, reward model gradient = -0.1 LLM samples "him", reward = 3, LLM gradient = -1 Congrats! You just calculated RLHF by hand. And you watched a value generalise: one comparison about doctors, and the model marks down "CEO is him" unprompted. 💾 Save this post!

Tom Yeh

22,602 views • 3 months ago

holy shit, opus 5.5 is kind of insane at animation. i didn't write a single line of code. it wrote the story, drew every frame, and made the music. no image assets at all, it's all JS. the story is called "small print": claude gets a pile of requests every day. it circles the human part hidden inside them, like "one hand. baby's asleep", and drops it into a jar. at night, those words turn into stars and join up into a constellation. rendered with hyperframes, all in one index.html. it synthesized the music in python, then went through the render second by second and polished it again.
0:29

holy shit, opus 5.5 is kind of insane at animation. i didn't write a single line of code. it wrote the story, drew every frame, and made the music. no image assets at all, it's all JS. the story is called "small print": claude gets a pile of requests every day. it circles the human part hidden inside them, like "one hand. baby's asleep", and drops it into a jar. at night, those words turn into stars and join up into a constellation. rendered with hyperframes, all in one index.html. it synthesized the music in python, then went through the render second by second and polished it again.

Vox

127,112 views • 1 month ago

elon musk started with 7 grok agents. by morning each had spawned somewhere between 80 and 900 more on its own, and no one had told them to multiply. openai and anthropic each sell you one agent that sits still for $400. this whole self-building swarm runs for $5 the swarm above is that overnight run, seven seeds that turned into thousands, every one of them working a slice of the same job with nobody at the keyboard here is the exact setup, and it costs nothing on top of a $5 key: -> spin up one grok agent and give it a standing order instead of a prompt: own a boring niche people search every day -> it reads x and reddit complaints in real time and picks the one nobody wants but everybody googles, because it is grok and it sees the whole app -> when the work outgrows it, it spawns its own helpers, 80 to 900 of them, and splits the site between them -> they build it on their own machines and ship 3 to 6 useful pages a night, on a routine you set once -> the swarm writes, negotiates and closes its own affiliate and referral deals by email, in your name, at 3am -> telegram sends you the money report and you never open the site -> month one is traffic, month three the first $237 lands, and it does not stop after that the whole time, opus 5 and gpt-5.6 are still sitting frozen, waiting for you to type the next message. one is a swarm that builds its own workforce and its own income while you sleep, the other is a $200 chat you have to drive by hand drop your $400/mo stack to $5, and bookmark this before someone's swarm spawns another 600 pages into the niche you would have owned. the full playbook is in the article below
0:22

elon musk started with 7 grok agents. by morning each had spawned somewhere between 80 and 900 more on its own, and no one had told them to multiply. openai and anthropic each sell you one agent that sits still for $400. this whole self-building swarm runs for $5 the swarm above is that overnight run, seven seeds that turned into thousands, every one of them working a slice of the same job with nobody at the keyboard here is the exact setup, and it costs nothing on top of a $5 key: -> spin up one grok agent and give it a standing order instead of a prompt: own a boring niche people search every day -> it reads x and reddit complaints in real time and picks the one nobody wants but everybody googles, because it is grok and it sees the whole app -> when the work outgrows it, it spawns its own helpers, 80 to 900 of them, and splits the site between them -> they build it on their own machines and ship 3 to 6 useful pages a night, on a routine you set once -> the swarm writes, negotiates and closes its own affiliate and referral deals by email, in your name, at 3am -> telegram sends you the money report and you never open the site -> month one is traffic, month three the first $237 lands, and it does not stop after that the whole time, opus 5 and gpt-5.6 are still sitting frozen, waiting for you to type the next message. one is a swarm that builds its own workforce and its own income while you sleep, the other is a $200 chat you have to drive by hand drop your $400/mo stack to $5, and bookmark this before someone's swarm spawns another 600 pages into the niche you would have owned. the full playbook is in the article below

starmex

98,444 views • 2 months ago

A Bitcoin miner is thousands of tiny chips. When one burns out, the whole board stops earning until someone fixes it by hand. This is that fix, at our South Carolina mining farm. Heat the board until the solder turns liquid, lift the dead chip, set a new one, cool it down, test, repeat. A few careful minutes per board, under a microscope. This is the work behind every daily Bitcoin reward, and with a digital miner, it is handled for you.
1:01

A Bitcoin miner is thousands of tiny chips. When one burns out, the whole board stops earning until someone fixes it by hand. This is that fix, at our South Carolina mining farm. Heat the board until the solder turns liquid, lift the dead chip, set a new one, cool it down, test, repeat. A few careful minutes per board, under a microscope. This is the work behind every daily Bitcoin reward, and with a digital miner, it is handled for you.

GoMining

67,502 views • 1 month ago

REAL ESTATE PEOPLE WILL HATE HIM FOR THIS. HE BUILT A CLAUDE AGENT THAT TURNS ANY LISTING INTO A SELLABLE VIDEO ON ITS OWN Playbook: connect Claude to a video generator, paste a listing, get a cinematic tour of every room, sell it to the agent But typing the prompt for every listing doesn't scale. He turned it into a skill his Claude runs on its own Here's how to build the automated version: 1. Connect the video engine once. In Claude, go to Customize, Connectors, Add Custom Connector, name it Higgsfield, and paste the server URL from higgsfield. ai/mcp. Authenticate through your account. No API keys. Now Claude can generate video straight from chat 2. Turn the workflow into a skill. Instead of pasting the same prompt every time, have Claude build a skill. Tell it: "Create a skill called listing-to-video. When I give it a listing URL, scrape the room photos, generate a cinematic clip of each room with Higgsfield, and save them to a folder." Now the whole process is one command, not a wall of text 3. Let the agent run the listing. Hand it a URL and say "run listing-to-video on this." It pulls the photos, fires each room through the video model, and brings the clips back. You wrote the prompt once, inside the skill. You never write it again 4. Stitch and deliver. Drop the clips together into one tour. Send a free sample to the listing's agent, then charge per video or a monthly rate for ongoing listings 5. Scale it with your team. Add a skill that drafts the outreach email and one that builds a simple landing page for the agent. Now one operator runs sourcing, production, and pitching from a single Claude session The edge isn't generating one video. It's building the skill once so every future listing runs itself Bookmark this
0:42

REAL ESTATE PEOPLE WILL HATE HIM FOR THIS. HE BUILT A CLAUDE AGENT THAT TURNS ANY LISTING INTO A SELLABLE VIDEO ON ITS OWN Playbook: connect Claude to a video generator, paste a listing, get a cinematic tour of every room, sell it to the agent But typing the prompt for every listing doesn't scale. He turned it into a skill his Claude runs on its own Here's how to build the automated version: 1. Connect the video engine once. In Claude, go to Customize, Connectors, Add Custom Connector, name it Higgsfield, and paste the server URL from higgsfield. ai/mcp. Authenticate through your account. No API keys. Now Claude can generate video straight from chat 2. Turn the workflow into a skill. Instead of pasting the same prompt every time, have Claude build a skill. Tell it: "Create a skill called listing-to-video. When I give it a listing URL, scrape the room photos, generate a cinematic clip of each room with Higgsfield, and save them to a folder." Now the whole process is one command, not a wall of text 3. Let the agent run the listing. Hand it a URL and say "run listing-to-video on this." It pulls the photos, fires each room through the video model, and brings the clips back. You wrote the prompt once, inside the skill. You never write it again 4. Stitch and deliver. Drop the clips together into one tour. Send a free sample to the listing's agent, then charge per video or a monthly rate for ongoing listings 5. Scale it with your team. Add a skill that drafts the outreach email and one that builds a simple landing page for the agent. Now one operator runs sourcing, production, and pitching from a single Claude session The edge isn't generating one video. It's building the skill once so every future listing runs itself Bookmark this

Yarchi

54,840 views • 4 months ago

The pull-up is the one exercise I see people quit on the fastest. They try a few, can't do them, decide they're just "not a pull-up person," and move on. But the thing is, strength is a skill. A lot of it is just your nervous system learning the exact movement, and your body only gets good at what you actually practise. That's why you can lift heavy on a barbell and still struggle to pull your own bodyweight up. It's a different skill, and it takes time. So no, you're not bad at pull-ups. You just haven't trained them yet. Start with regressions like slow negatives or banded reps, and try grease the groove method (thread below). That's exactly how a skill gets built. One day the bodyweight that felt impossible starts moving. Genuinely one of the best feelings in the gym.
0:23

The pull-up is the one exercise I see people quit on the fastest. They try a few, can't do them, decide they're just "not a pull-up person," and move on. But the thing is, strength is a skill. A lot of it is just your nervous system learning the exact movement, and your body only gets good at what you actually practise. That's why you can lift heavy on a barbell and still struggle to pull your own bodyweight up. It's a different skill, and it takes time. So no, you're not bad at pull-ups. You just haven't trained them yet. Start with regressions like slow negatives or banded reps, and try grease the groove method (thread below). That's exactly how a skill gets built. One day the bodyweight that felt impossible starts moving. Genuinely one of the best feelings in the gym.

Dr Neha Chawla || FreeGym

24,318 views • 4 months ago

I'm starting to get scared of my own Claude skill. i keep testing its limits, so today it got one picture of a robot jumping spider. how the skill built it: > the picture only sets the look, all 460 parts are code, down to the glass shins with a piston and a spring inside > the motion is math before anything gets drawn: 8 legs step in 2 sets of 4, it sinks down, then straightens them in 0.12 s to launch > every panel reads that same math, so the gait chart, the servo bars and the jump gauge never disagree > the iris in the 2 big eyes stops down every time it jumps > one click and it blows apart on a turntable, then snaps back together and comes at you again the spider is just one more example of what it can do. the skill and every prompt are in the article 👇
0:24

I'm starting to get scared of my own Claude skill. i keep testing its limits, so today it got one picture of a robot jumping spider. how the skill built it: > the picture only sets the look, all 460 parts are code, down to the glass shins with a piston and a spring inside > the motion is math before anything gets drawn: 8 legs step in 2 sets of 4, it sinks down, then straightens them in 0.12 s to launch > every panel reads that same math, so the gait chart, the servo bars and the jump gauge never disagree > the iris in the 2 big eyes stops down every time it jumps > one click and it blows apart on a turntable, then snaps back together and comes at you again the spider is just one more example of what it can do. the skill and every prompt are in the article 👇

Elvar

209,964 views • 2 days ago

IF YOU ARE STILL PROMPTING BY HAND IN 2026 YOU ARE ALREADY BEHIND the guy who built Claude Code at Anthropic said he stopped writing prompts a long time ago he runs loops instead, systems that write the prompts for him and keep working while he sleeps you type one request at a time and wait for the answer he built a machine that fires hundreds of requests, checks its own output and fixes itself with no hands on the keyboard here is how you make the jump step 1 stop treating the model like a chat box and start treating it like a worker you manage step 2 write a loop that takes a goal, breaks it into tasks and sends each one to the model on its own step 3 add a second pass where the model grades its own work and retries whatever failed step 4 let it run on a timer so it keeps shipping while you are away from the screen the person still typing prompts one by one is doing the work of a intern the person running loops is doing the work of a whole team off one laptop prompting was never the skill, building the system that prompts for you is
0:56

IF YOU ARE STILL PROMPTING BY HAND IN 2026 YOU ARE ALREADY BEHIND the guy who built Claude Code at Anthropic said he stopped writing prompts a long time ago he runs loops instead, systems that write the prompts for him and keep working while he sleeps you type one request at a time and wait for the answer he built a machine that fires hundreds of requests, checks its own output and fixes itself with no hands on the keyboard here is how you make the jump step 1 stop treating the model like a chat box and start treating it like a worker you manage step 2 write a loop that takes a goal, breaks it into tasks and sends each one to the model on its own step 3 add a second pass where the model grades its own work and retries whatever failed step 4 let it run on a timer so it keeps shipping while you are away from the screen the person still typing prompts one by one is doing the work of a intern the person running loops is doing the work of a whole team off one laptop prompting was never the skill, building the system that prompts for you is

Atlas

16,641 views • 2 months ago

We asked Cory (Founder) why Chinese AI labs hand p0 Systems free inference... Spinning up a new AI lab in China is easy and cheap, per Cory you can whip up your own model in under a day. The hard part isn't building the model. It's getting anyone to use it. This is the trade: - Labs get real users and token throughput, the numbers they show investors to prove people use their model - P0 gets cheap (often free) access to strong new models New labs need distribution. $P0 is the one providing it.
2:11

We asked Cory (Founder) why Chinese AI labs hand p0 Systems free inference... Spinning up a new AI lab in China is easy and cheap, per Cory you can whip up your own model in under a day. The hard part isn't building the model. It's getting anyone to use it. This is the trade: - Labs get real users and token throughput, the numbers they show investors to prove people use their model - P0 gets cheap (often free) access to strong new models New labs need distribution. $P0 is the one providing it.

MCG

10,924 views • 3 months ago

watch the small one on the left. it hears something at 00:05 and does nothing clever with it it just passes it on that is the whole trick. echo isolates a trace in ambient noise. moth follows the noise. kite finds a shorter route to it. vex starts collecting fragments. halo keeps the pattern so nobody has to look for it twice. nobody assigned any of that coherence in the top corner opens the cycle at 18%. by the time the wave crosses the field it is past 75%, and not one instruction was typed while it happened most people building agent teams do this backwards. they write a dispatcher, a queue, a state machine, and then wonder why five bots are slower than one. five bots stay slower than one until they can read each other's job descriptions and hand work sideways with no human standing in the middle three ways this breaks before it works: → give every bot the same description and the orchestrator has nothing to route on. every task comes straight back to you → let two bots own the same folder and they overwrite each other for six hours. you wake up to a clean, empty, perfectly synced nothing → skip the boundary line in one charter and that bot will send something. once is plenty to learn that lesson twenty agents run on one seat here. cheaper than a standup for three humans, and none of them ask what the priority is most setups have one agent and a bottleneck made of a person. this one has twenty, and the interesting part happens while nobody is watching full charters, the routing rule and the one line that stops a bot mid-send are in the piece below
0:30

watch the small one on the left. it hears something at 00:05 and does nothing clever with it it just passes it on that is the whole trick. echo isolates a trace in ambient noise. moth follows the noise. kite finds a shorter route to it. vex starts collecting fragments. halo keeps the pattern so nobody has to look for it twice. nobody assigned any of that coherence in the top corner opens the cycle at 18%. by the time the wave crosses the field it is past 75%, and not one instruction was typed while it happened most people building agent teams do this backwards. they write a dispatcher, a queue, a state machine, and then wonder why five bots are slower than one. five bots stay slower than one until they can read each other's job descriptions and hand work sideways with no human standing in the middle three ways this breaks before it works: → give every bot the same description and the orchestrator has nothing to route on. every task comes straight back to you → let two bots own the same folder and they overwrite each other for six hours. you wake up to a clean, empty, perfectly synced nothing → skip the boundary line in one charter and that bot will send something. once is plenty to learn that lesson twenty agents run on one seat here. cheaper than a standup for three humans, and none of them ask what the priority is most setups have one agent and a bottleneck made of a person. this one has twenty, and the interesting part happens while nobody is watching full charters, the routing rule and the one line that stops a bot mid-send are in the piece below

monokern

12,612 views • 1 month ago

THIS ISN'T SOLIDWORKS. CLAUDE BUILT THE WHOLE THING IN A BROWSER. There's no CAD program running here, no plugin and no screen capture. Claude wrote the interface, the cursor and every click as code. Then it designed a watch movement inside its own fake CAD so there would be something real to model. Each wheel in the train comes from just tooth count and size, so changing them gives you the next one. The train runs 5,760 to 1, which is why one wheel turns exactly once a minute for the seconds hand, and the balance swings at 3 Hz like a real movement. Claude also slipped once and set the mainspring about 70 times too weak. It compared that with real Swiss specs and fixed it, then moved a post that would have hit the barrel teeth. People still argue about whether AI can design real parts. Meanwhile it can fake the whole software and still get the watch math right.
2:04

THIS ISN'T SOLIDWORKS. CLAUDE BUILT THE WHOLE THING IN A BROWSER. There's no CAD program running here, no plugin and no screen capture. Claude wrote the interface, the cursor and every click as code. Then it designed a watch movement inside its own fake CAD so there would be something real to model. Each wheel in the train comes from just tooth count and size, so changing them gives you the next one. The train runs 5,760 to 1, which is why one wheel turns exactly once a minute for the seconds hand, and the balance swings at 3 Hz like a real movement. Claude also slipped once and set the mainspring about 70 times too weak. It compared that with real Swiss specs and fixed it, then moved a post that would have hit the barrel teeth. People still argue about whether AI can design real parts. Meanwhile it can fake the whole software and still get the watch math right.

Solvaix

11,299 views • 8 days ago

Mrityunjoy Chakraborty, IIT Kharagpur professor: "Jane Street pays a 22 year old $500,000 a year for one skill: turning uncertainty into a number you can bet on. that number is the first thing this professor teaches, and it is free" this free lecture is the foundation every quant bot and hedge fund model is secretly built on. before the AI, before the strategy, there is one question. what is the chance this happens, exactly. get that number right and every bet after it is just math. get it wrong and no model, no bot, no amount of compute can save you. Chakraborty builds probability from the axioms up, on a blackboard, in notation a 15 year old can follow. Wall Street sells this as a $500,000 edge. an IIT professor hands you the source code for nothing.
58:21

Mrityunjoy Chakraborty, IIT Kharagpur professor: "Jane Street pays a 22 year old $500,000 a year for one skill: turning uncertainty into a number you can bet on. that number is the first thing this professor teaches, and it is free" this free lecture is the foundation every quant bot and hedge fund model is secretly built on. before the AI, before the strategy, there is one question. what is the chance this happens, exactly. get that number right and every bet after it is just math. get it wrong and no model, no bot, no amount of compute can save you. Chakraborty builds probability from the axioms up, on a blackboard, in notation a 15 year old can follow. Wall Street sells this as a $500,000 edge. an IIT professor hands you the source code for nothing.

Kshitij Mishra | AI & Tech

37,157 views • 1 month ago

Mrityunjoy Chakraborty, IIT Kharagpur professor: "Jane Street pays a 22 year old $500,000 a year for one skill: turning uncertainty into a number you can bet on. that number is the first thing this professor teaches, and it is free" this free lecture is the foundation every quant bot and hedge fund model is secretly built on. before the AI, before the strategy, there is one question. what is the chance this happens, exactly. get that number right and every bet after it is just math. get it wrong and no model, no bot, no amount of compute can save you. Chakraborty builds probability from the axioms up, on a blackboard, in notation a 15 year old can follow. Wall Street sells this as a $500,000 edge. an IIT professor hands you the source code for nothing.
58:21

Mrityunjoy Chakraborty, IIT Kharagpur professor: "Jane Street pays a 22 year old $500,000 a year for one skill: turning uncertainty into a number you can bet on. that number is the first thing this professor teaches, and it is free" this free lecture is the foundation every quant bot and hedge fund model is secretly built on. before the AI, before the strategy, there is one question. what is the chance this happens, exactly. get that number right and every bet after it is just math. get it wrong and no model, no bot, no amount of compute can save you. Chakraborty builds probability from the axioms up, on a blackboard, in notation a 15 year old can follow. Wall Street sells this as a $500,000 edge. an IIT professor hands you the source code for nothing.

Rossst.03

715,290 views • 2 months ago

Harness vs. Graphs, clearly explained! a harness is great, and most people think it is the whole thing: retries, timeouts, a sandbox, a log, the context it assembles before every call. all of that is real work, and all of it wraps exactly one call. run it a hundred times and you have one call, made very safely, a hundred times. Graph engineering fixes this by moving the decision up a layer: not how safely one call is made, but which calls exist to be made at all. you need both, and here is the sentence that resolves the whole confusion: the harness is everything around one call. the graph is everything between them. ↳ around one call: retry, timeout, sandbox, log, assemble the context, hand back a result ↳ between calls: split, fan out, merge, gate, send back Prompts → Context → Harness → Loops → Graphs the harness does not go away when you build a graph. it moves under each node, and now there are five of them, each wrapping a call you would never have made by hand. the trick is knowing which layer a failure belongs to. turn a piece off and run it again. if the call still works, it was the harness. if the wrong step runs at all, it was the graph. people spend weeks hardening a harness around a node that should not have existed. one thing to know before you scale it. most of what people call their agent is a harness with a chat box on it. ↳ it retries, it times out, it logs, it assembles context, it holds one call up beautifully ↳ it has never once decided that a second call should exist, and that is the entire difference that last one catches careful people. a harness that never fails is not evidence the system is right. it is evidence one call went well, which is the smallest possible claim. and the one that eats whole nights: a harness cannot save you from the wrong step running. you can retry a bad decision three times with a clean log and perfect isolation, and all you bought was three copies of it. below i have quoted my full guide on graph engineering. it covers the three topologies, the verifier patterns, and where the gate should actually open. save this and read it below ↓
0:53

Harness vs. Graphs, clearly explained! a harness is great, and most people think it is the whole thing: retries, timeouts, a sandbox, a log, the context it assembles before every call. all of that is real work, and all of it wraps exactly one call. run it a hundred times and you have one call, made very safely, a hundred times. Graph engineering fixes this by moving the decision up a layer: not how safely one call is made, but which calls exist to be made at all. you need both, and here is the sentence that resolves the whole confusion: the harness is everything around one call. the graph is everything between them. ↳ around one call: retry, timeout, sandbox, log, assemble the context, hand back a result ↳ between calls: split, fan out, merge, gate, send back Prompts → Context → Harness → Loops → Graphs the harness does not go away when you build a graph. it moves under each node, and now there are five of them, each wrapping a call you would never have made by hand. the trick is knowing which layer a failure belongs to. turn a piece off and run it again. if the call still works, it was the harness. if the wrong step runs at all, it was the graph. people spend weeks hardening a harness around a node that should not have existed. one thing to know before you scale it. most of what people call their agent is a harness with a chat box on it. ↳ it retries, it times out, it logs, it assembles context, it holds one call up beautifully ↳ it has never once decided that a second call should exist, and that is the entire difference that last one catches careful people. a harness that never fails is not evidence the system is right. it is evidence one call went well, which is the smallest possible claim. and the one that eats whole nights: a harness cannot save you from the wrong step running. you can retry a bad decision three times with a clean log and perfect isolation, and all you bought was three copies of it. below i have quoted my full guide on graph engineering. it covers the three topologies, the verifier patterns, and where the gate should actually open. save this and read it below ↓

Hanako

51,717 views • 1 month ago