Загрузка видео...

Не удалось загрузить видео

Возникла проблема при загрузке этого видео. Это может быть связано с временными проблемами сети или видео может быть недоступно.

На главную

An hour of lab evals catches a model that doesn't work. It won't catch one that fails once every two hundred trials, or degrades over a week, or runs fine on this robot and badly on the one beside it. So the eval moved to where the work is.... show more

Dyna Robotics

11,190 subscribers

10,964 просмотров • 17 дней назад •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

For robotics researchers, evaluation can be the slowest part of their research loop. Manifold makes it the fastest. Introducing Manifold. The platform for accelerating robotics research with failure analysis that explains itself. A month of experiments in a week. Simulate a thousand scenarios in 30 minutes. Dense labels on every atomic sub-goal and action. Agents that cluster failure modes across every rollout. The robotics research loop that moves at the speed of compute. One eval harness for every benchmark, simulator, or world model. Native support for NVIDIA’s Isaac Lab Arena.

For robotics researchers, evaluation can be the slowest part of their research loop. Manifold makes it the fastest. Introducing Manifold. The platform for accelerating robotics research with failure analysis that explains itself. A month of experiments in a week. Simulate a thousand scenarios in 30 minutes. Dense labels on every atomic sub-goal and action. Agents that cluster failure modes across every rollout. The robotics research loop that moves at the speed of compute. One eval harness for every benchmark, simulator, or world model. Native support for NVIDIA’s Isaac Lab Arena.

skara

24,272 просмотров • 1 месяц назад

It was a great week for innovation across the model ecosystem. We're bringing these new models into Copilot, enabling it to take on increasingly complex work, from quick questions to delegated tasks and, increasingly, complete long-running jobs through Autopilots. Here's one fun example: using an Opal-powered Autopilot running on a secure Windows 365 Cloud PC, Copilot goes through a month of trail cam footage, finds every animal sighting, creates a highlight reel of the best clips, labels each one with the camera, date, and species, catalogs every sighting in a spreadsheet, builds a PowerPoint summarizing the findings, and shares everything in Teams for colleagues to review. This is the next frontier for Copilot: software that doesn't just assist with work, but can own and complete entire tasks that unfold over hours or days.

It was a great week for innovation across the model ecosystem. We're bringing these new models into Copilot, enabling it to take on increasingly complex work, from quick questions to delegated tasks and, increasingly, complete long-running jobs through Autopilots. Here's one fun example: using an Opal-powered Autopilot running on a secure Windows 365 Cloud PC, Copilot goes through a month of trail cam footage, finds every animal sighting, creates a highlight reel of the best clips, labels each one with the camera, date, and species, catalogs every sighting in a spreadsheet, builds a PowerPoint summarizing the findings, and shares everything in Teams for colleagues to review. This is the next frontier for Copilot: software that doesn't just assist with work, but can own and complete entire tasks that unfold over hours or days.

Satya Nadella

336,035 просмотров • 7 дней назад

$Roboflow charges you for models you have to verify yourself. Introducing Score Studio: the decentralised alternative where every model arrives already proven and ready to work in real world conditions. Every model on the platform is battle-tested through SN44's validation mechanism before you ever touch it. Miners compete, validators score, only the winners survive. You get the output of a permanent open competition, at a fraction of the price. Datasets, annotation, training, deployment. All of it cheaper. All of it validated, accurate and efficient. Need a model that doesn't exist yet? Launch your own model inquiry from the Studio and put the network to work on it. Pay in USD or SN44, straight at subnet level. We're putting the final touches on it now.$

Roboflow charges you for models you have to verify yourself. Introducing Score Studio: the decentralised alternative where every model arrives already proven and ready to work in real world conditions. Every model on the platform is battle-tested through SN44's validation mechanism before you ever touch it. Miners compete, validators score, only the winners survive. You get the output of a permanent open competition, at a fraction of the price. Datasets, annotation, training, deployment. All of it cheaper. All of it validated, accurate and efficient. Need a model that doesn't exist yet? Launch your own model inquiry from the Studio and put the network to work on it. Pay in USD or SN44, straight at subnet level. We're putting the final touches on it now.

Score

35,258 просмотров • 1 месяц назад

Someone just posted the full blueprint for an AI swarm that does the job of a 200-person quant research team. Six agents. Running 24/7. Finding brand-new alpha while you sleep. Citadel needs 100 PhDs to do this. Two Sigma needs 200. This does it with six bots and one laptop. Two ways to play this - spend a weekend building your own swarm, or copy the wallet of one that's already up $2M: Boris Cherny runs Claude Code at Anthropic. Two weeks ago he said: "I don't prompt Claude anymore. I have loops running that prompt Claude. My job is to write loops" Alpha research is just a pipeline. So instead of sitting in it, you hand each stage to its own agent: > one reads every new research paper overnight and pulls out the trade idea > one builds the features and cleans the data > one backtests it over 20 years, costs and slippage included > one runs the hard stats and kills anything overfit > one checks it still works in every market regime > one strips out plain momentum and value to see if any real edge is left Each of those six is a job a fund pays a $600,000-a-year quant to do. He runs all six for the price of an API bill. The rule that makes it work: the agent that builds a signal never gets to approve it. A separate, stronger agent tries to kill it first. Whatever survives all six by morning is real, new alpha. One trader's already running this exact swarm on Polymarket. That $2M wallet is public, every trade on-chain. The full build is in the post below - six agents, the tool that runs them, and the five mistakes that kill most people. Bookmark & read this before it's buried.

cvxv666

103,734 просмотров • 2 месяцев назад

Wayne Gretzky shares the advice he gives to every young athlete. "The commitment and the work ethic is worth it." "My dad was a big believer that the harder you work as a kid, the more fun you're going to have as an adult. And the more fun you have as a kid, the harder you're going to work as an adult." Read that again. It's one of the truest things ever said about success. Work now, play later. Or play now and work forever. "The commitment and the passion for what you're doing is so worth it at the end of the day. Just believe each and every day - it's really, really worth it." Commitment is a choice. It's choosing long-term growth over short-term comfort. "I wouldn't trade my life in for anything in the world." Outcomes and rewards require sacrifice. The key question is: what price am I willing to pay (🎥Be Better Off Show)

Wayne Gretzky shares the advice he gives to every young athlete. "The commitment and the work ethic is worth it." "My dad was a big believer that the harder you work as a kid, the more fun you're going to have as an adult. And the more fun you have as a kid, the harder you're going to work as an adult." Read that again. It's one of the truest things ever said about success. Work now, play later. Or play now and work forever. "The commitment and the passion for what you're doing is so worth it at the end of the day. Just believe each and every day - it's really, really worth it." Commitment is a choice. It's choosing long-term growth over short-term comfort. "I wouldn't trade my life in for anything in the world." Outcomes and rewards require sacrifice. The key question is: what price am I willing to pay (🎥Be Better Off Show)

Coach AJ 🎯 Mental Fitness

50,797 просмотров • 24 дней назад

Wayne Gretzky shares the advice he gives to every young athlete. "The commitment and the work ethic is worth it." "My dad was a big believer that the harder you work as a kid, the more fun you're going to have as an adult. And the more fun you have as a kid, the harder you're going to work as an adult." Read that again. It's one of the truest things ever said about success. Work now, play later. Or play now and work forever. "The commitment and the passion for what you're doing is so worth it at the end of the day. Just believe each and every day - it's really, really worth it." Commitment is a choice. It's choosing long-term growth over short-term comfort. "I wouldn't trade my life in for anything in the world." Outcomes and rewards require sacrifice. The key question is: what price am I willing to pay? (🎥 Be Better Off Show)

Wayne Gretzky shares the advice he gives to every young athlete. "The commitment and the work ethic is worth it." "My dad was a big believer that the harder you work as a kid, the more fun you're going to have as an adult. And the more fun you have as a kid, the harder you're going to work as an adult." Read that again. It's one of the truest things ever said about success. Work now, play later. Or play now and work forever. "The commitment and the passion for what you're doing is so worth it at the end of the day. Just believe each and every day - it's really, really worth it." Commitment is a choice. It's choosing long-term growth over short-term comfort. "I wouldn't trade my life in for anything in the world." Outcomes and rewards require sacrifice. The key question is: what price am I willing to pay? (🎥 Be Better Off Show)

Coach AJ 🎯 Mental Fitness

181,918 просмотров • 7 месяцев назад

A DOCTOR BUILT AN AI-QUERYABLE DATABASE OF EVERY DISEASE HE KNOWS IN OBSIDIAN. THE SAME STACK CAN RUN AN ENTIRE COMPANY ON ONE PERSON'S BRAIN Zoomed out, his vault looks like a constellation. Every disease linked to its symptoms, every symptom to the conditions it shows up in, every treatment to what it treats. Adding a new case is one note that wires itself into the rest. It's a personal database, and the structure has nothing to do with medicine. It's just nodes and links. Once you see that, the same setup runs anything: > a startup that needs to remember every customer, decision, and meeting; > a school whose curriculum links to every lecture and student question; > a business where SOPs, clients, and vendors all sit in one queryable graph. Drop Claude on top, and the vault stops being a map. It becomes something you can ask, in plain English, across everything you've ever written down. Full Claude + Obsidian build in the article below. Bookmark this

A DOCTOR BUILT AN AI-QUERYABLE DATABASE OF EVERY DISEASE HE KNOWS IN OBSIDIAN. THE SAME STACK CAN RUN AN ENTIRE COMPANY ON ONE PERSON'S BRAIN Zoomed out, his vault looks like a constellation. Every disease linked to its symptoms, every symptom to the conditions it shows up in, every treatment to what it treats. Adding a new case is one note that wires itself into the rest. It's a personal database, and the structure has nothing to do with medicine. It's just nodes and links. Once you see that, the same setup runs anything: > a startup that needs to remember every customer, decision, and meeting; > a school whose curriculum links to every lecture and student question; > a business where SOPs, clients, and vendors all sit in one queryable graph. Drop Claude on top, and the vault stops being a map. It becomes something you can ask, in plain English, across everything you've ever written down. Full Claude + Obsidian build in the article below. Bookmark this

Yarchi

32,594 просмотров • 2 месяцев назад

$The AI that will replace lawyers was just built for $40 million, and every lawyer alive helped train it themselves. Frontier labs spend billions to reach the top. But here we have a 150-year-old legal publisher who got there for $40 million. Thomson Reuters owns Westlaw. Every lawyer in America has paid to use it. It's the database you search to find the case that wins your argument, and the profession has run on it for generations. And they just turned it into a weapon. They took a free open-weight model off the internet. Then they poured Westlaw into it: 40,000 databases, 150 years of legal writing, all curated by the profession itself. The result beats some frontier models at legal work. And the $40 million was mostly talent and two years of work. The final training run cost about $450,000. That's the machine aimed at a trillion-dollar profession, trained for the price of one associate's salary. Every ruling a lawyer ever wrote, every brief that got published, every annotation an editor ever added to Westlaw is the fuel. The archive that lawyers spent a century building is the exact thing now being trained to do their jobs. The model is called Thomson. It runs at a fraction of the cost of a normal frontier model, because it's small and specialized instead of giant and general. You don't need a $500 billion supercluster to replace a knowledge worker. You need their field's best data and about $40 million. So think about who else is sitting on decades of proprietary data: - Every medical publisher - Every accounting firm - Every engineering standards body - Every company that ever made its experts write down what they know The tool that ends the billable hour just proved the playbook. And the playbook works on almost every white-collar profession that runs on a specialized archive. But also keep in mind that Thomson Reuters owns this content outright. No customer's private files went into it, and they say the goal is to make lawyers faster, not to fire them (that's what they say). Today Thomson only does one job inside CoCounsel: Reviewing stacks of documents and sorting them into tables. It isn't arguing cases yet. A better research tool has always meant one lawyer does the work of three, not that the other two vanish. The trouble is what "faster" has always meant in practice. When one lawyer can suddenly do the work of five, a firm doesn't keep five lawyers. It keeps one and bills the same. The junior associate who used to do the research is the cost that DISAPPEARS. Big Law was built on armies of associates grinding through documents by the hour. That grind is the exact job Thomson does now in seconds. So the profession that spent 150 years filling the archive just watched that archive get turned into the thing that makes most of them unnecessary. Who does it next and in what field?$

The AI that will replace lawyers was just built for $40 million, and every lawyer alive helped train it themselves. Frontier labs spend billions to reach the top. But here we have a 150-year-old legal publisher who got there for $40 million. Thomson Reuters owns Westlaw. Every lawyer in America has paid to use it. It's the database you search to find the case that wins your argument, and the profession has run on it for generations. And they just turned it into a weapon. They took a free open-weight model off the internet. Then they poured Westlaw into it: 40,000 databases, 150 years of legal writing, all curated by the profession itself. The result beats some frontier models at legal work. And the $40 million was mostly talent and two years of work. The final training run cost about $450,000. That's the machine aimed at a trillion-dollar profession, trained for the price of one associate's salary. Every ruling a lawyer ever wrote, every brief that got published, every annotation an editor ever added to Westlaw is the fuel. The archive that lawyers spent a century building is the exact thing now being trained to do their jobs. The model is called Thomson. It runs at a fraction of the cost of a normal frontier model, because it's small and specialized instead of giant and general. You don't need a $500 billion supercluster to replace a knowledge worker. You need their field's best data and about $40 million. So think about who else is sitting on decades of proprietary data: - Every medical publisher - Every accounting firm - Every engineering standards body - Every company that ever made its experts write down what they know The tool that ends the billable hour just proved the playbook. And the playbook works on almost every white-collar profession that runs on a specialized archive. But also keep in mind that Thomson Reuters owns this content outright. No customer's private files went into it, and they say the goal is to make lawyers faster, not to fire them (that's what they say). Today Thomson only does one job inside CoCounsel: Reviewing stacks of documents and sorting them into tables. It isn't arguing cases yet. A better research tool has always meant one lawyer does the work of three, not that the other two vanish. The trouble is what "faster" has always meant in practice. When one lawyer can suddenly do the work of five, a firm doesn't keep five lawyers. It keeps one and bills the same. The junior associate who used to do the research is the cost that DISAPPEARS. Big Law was built on armies of associates grinding through documents by the hour. That grind is the exact job Thomson does now in seconds. So the profession that spent 150 years filling the archive just watched that archive get turned into the thing that makes most of them unnecessary. Who does it next and in what field?

Ricardo

26,901 просмотров • 18 дней назад

It is a new year! With it comes a challenge every week to reflect on one positive thing that happened to you and write it down. At the end of the year, we can reflect on all the good times. What's one good thing that happened to you this week? Elon Musk Gamer Linda Yaccarino

It is a new year! With it comes a challenge every week to reflect on one positive thing that happened to you and write it down. At the end of the year, we can reflect on all the good times. What's one good thing that happened to you this week? Elon Musk Gamer Linda Yaccarino

Adam Lowisz X Meetup 🇺🇸🇵🇱🇪🇺🇬🇧🇺🇦

472,217 просмотров • 1 год назад

Apple spending $10B across 10 years to build a new product is equivalent to spending $83M per month or $3M a day. This was NOT just a side project, it was a MAJOR project that was supposed to build the first Apple Car to take on Tesla head on. This week the largest company in the world waved the white flag and gave up on it. Every Tesla you see on the road is proof that the company & team accomplished the impossible, which is creating an EV that autonomously drives itself, one that is produced at scale & at a profit, and one that people absolutely love.

Apple spending $10B across 10 years to build a new product is equivalent to spending $83M per month or $3M a day. This was NOT just a side project, it was a MAJOR project that was supposed to build the first Apple Car to take on Tesla head on. This week the largest company in the world waved the white flag and gave up on it. Every Tesla you see on the road is proof that the company & team accomplished the impossible, which is creating an EV that autonomously drives itself, one that is produced at scale & at a profit, and one that people absolutely love.

Teslaconomics

330,663 просмотров • 2 лет назад

It’s one of my favorite days of the year… Happy #InternationalWomensDay! #IWD23 is a day to reflect on the work of women throughout history and the work that continues today. It’s also a day to celebrate the women in your own life—like your mom, your sister, or your friends.

It’s one of my favorite days of the year… Happy #InternationalWomensDay! #IWD23 is a day to reflect on the work of women throughout history and the work that continues today. It’s also a day to celebrate the women in your own life—like your mom, your sister, or your friends.

Anita Zaidi

672,258 просмотров • 3 лет назад

Tim Robbins says Clint Eastwood shot Mystic River with one take per setup and had the cast out by 2 PM every day "There's not a lot of bells and whistles on the set. There's no, 'Rolling!' and, 'Action!' It's all just very quiet, very zen-like from the moment you walk on the set" "And you also know, as Clint has told you, 'We're not gonna do a lot of takes. We're gonna do one take of each setup'" "That's what you get. You get one shot at each setup. That means you have three chances of playing the scene, one in the wide, one in the mid, and one in the tight. That's it, and personally, I did a second take on maybe two setups" "That's the discipline going in. And so the night before, as actors, we're rehearsing. We're ready. We're getting ready. We know every word perfectly" "We're not gonna go to the set and say, 'Hey, Clint, I think about this. What about if this happened?' That never happened" "So we were in at 9 AM and out by 2 PM every day. It was beautiful. It was efficient"

Tim Robbins says Clint Eastwood shot Mystic River with one take per setup and had the cast out by 2 PM every day "There's not a lot of bells and whistles on the set. There's no, 'Rolling!' and, 'Action!' It's all just very quiet, very zen-like from the moment you walk on the set" "And you also know, as Clint has told you, 'We're not gonna do a lot of takes. We're gonna do one take of each setup'" "That's what you get. You get one shot at each setup. That means you have three chances of playing the scene, one in the wide, one in the mid, and one in the tight. That's it, and personally, I did a second take on maybe two setups" "That's the discipline going in. And so the night before, as actors, we're rehearsing. We're ready. We're getting ready. We know every word perfectly" "We're not gonna go to the set and say, 'Hey, Clint, I think about this. What about if this happened?' That never happened" "So we were in at 9 AM and out by 2 PM every day. It was beautiful. It was efficient"

Casper

98,044 просмотров • 8 дней назад

JUST IN: Dyna Robotics just published one of the most important research papers in robotics this year. It could fundamentally change how robot foundation models are trained. A scaling law that transfers from human video to robot performance. Dyna-2 is out and it's 🔥 Here's what that means in plain terms. Dyna-2 was pre-trained on ONE MILLION hours of egocentric human video, 170 years of continuous human experience, cooking, folding, assembling, cleaning. And as that human data scaled, robot performance improved. Predictably. Monotonically. Across 39 tasks on two different robot embodiments the model had never seen. → 1,000 hours pre-training → 20% normalised task performance → 10,000 hours → 28% → 100,000 hours → 45% → 1,000,000 hours → 53% Human video exists at effectively unlimited scale. Every cook, every factory worker, every craftsperson wearing a camera is generating training data for future robots. But the finding that stunned even the researchers, world modeling is what makes the transfer work. A model trained to predict future video AND actions massively outperforms one trained on actions alone. Video is the new scaling axis for robotics. One more jaw-dropping data point. 13 minutes of teleoperation data was enough to fine-tune Dyna-2 to open a bottle cap using two five-fingered robot hands. The robots are coming, and they're learning from us directly :D Read more here: Congrats Jason Ma and team! ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

JUST IN: Dyna Robotics just published one of the most important research papers in robotics this year. It could fundamentally change how robot foundation models are trained. A scaling law that transfers from human video to robot performance. Dyna-2 is out and it's 🔥 Here's what that means in plain terms. Dyna-2 was pre-trained on ONE MILLION hours of egocentric human video, 170 years of continuous human experience, cooking, folding, assembling, cleaning. And as that human data scaled, robot performance improved. Predictably. Monotonically. Across 39 tasks on two different robot embodiments the model had never seen. → 1,000 hours pre-training → 20% normalised task performance → 10,000 hours → 28% → 100,000 hours → 45% → 1,000,000 hours → 53% Human video exists at effectively unlimited scale. Every cook, every factory worker, every craftsperson wearing a camera is generating training data for future robots. But the finding that stunned even the researchers, world modeling is what makes the transfer work. A model trained to predict future video AND actions massively outperforms one trained on actions alone. Video is the new scaling axis for robotics. One more jaw-dropping data point. 13 minutes of teleoperation data was enough to fine-tune Dyna-2 to open a bottle cap using two five-fingered robot hands. The robots are coming, and they're learning from us directly :D Read more here: Congrats Jason Ma and team! ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

Lukas Ziegler

23,576 просмотров • 1 месяц назад

Marauder on-water testing is in full swing. The team took Marauder from initial design to on-water trials in under 12 months. Now we're running it on the water 12+ hours a day, seven days a week, proving out the autonomy at range and in conditions that don't cooperate. We're pushing the system hard - it's the same bar we've held on every vessel we've fielded, and we're holding it here. We're putting in the work that matters and we're glad to be doing it.

Marauder on-water testing is in full swing. The team took Marauder from initial design to on-water trials in under 12 months. Now we're running it on the water 12+ hours a day, seven days a week, proving out the autonomy at range and in conditions that don't cooperate. We're pushing the system hard - it's the same bar we've held on every vessel we've fielded, and we're holding it here. We're putting in the work that matters and we're glad to be doing it.

Saronic

72,336 просмотров • 1 месяц назад

Good morning you gorgeous little pumpkins of loveliness. Look at all those beautiful little smiling faces of the commuters on London Bridge on this fine Monday morning as I cycle across. Every one of them excited to start a new week. Have a great one :)

Good morning you gorgeous little pumpkins of loveliness. Look at all those beautiful little smiling faces of the commuters on London Bridge on this fine Monday morning as I cycle across. Every one of them excited to start a new week. Have a great one :)

Alastair Hilton

83,421 просмотров • 7 месяцев назад

Engram cofounder Jack Morris just raised $98M to build a new type of AI. He says models don't need to get smarter over time. Instead, they just need to know you better and better over time. Jack describes what he's building: "Our product is a new type of AI. We have a pretty different vision from a lot of the frontier labs, which are working on one model per lab, and trying to make that model smarter every month." "There's another way to think about it, which is that the model doesn't need to get smarter every month. It needs to know you better." "So we're working on a whole different stack, which is a way to train models that train themselves to know your world better and adjust to the things that you say." "So: new ways of training, new ways of running the models."

Engram cofounder Jack Morris just raised $98M to build a new type of AI. He says models don't need to get smarter over time. Instead, they just need to know you better and better over time. Jack describes what he's building: "Our product is a new type of AI. We have a pretty different vision from a lot of the frontier labs, which are working on one model per lab, and trying to make that model smarter every month." "There's another way to think about it, which is that the model doesn't need to get smarter every month. It needs to know you better." "So we're working on a whole different stack, which is a way to train models that train themselves to know your world better and adjust to the things that you say." "So: new ways of training, new ways of running the models."

TBPN

70,552 просмотров • 2 месяцев назад

Happy Earth Day 🌍 Today is a reminder of this extraordinary gift we've been given! This extraordinary Blue Planet we all call home, floating through the vast universe, carrying every ocean, every desert, every mountain, every forest. Every breath of life we have ever known. ✨ We were so moved by the historic new images of Earth from deep space that NASA shared from their Artemis II mission. The sheer MIRACLE of it all fills me with a deep, profound gratitude AND Serves as a powerful reminder that we each are called to be stewards of this precious miracle of life. The quiet, sacred web of life that breathes through us ALL. And let us also remember to be good stewards of EACH OTHER as we share this journey — every single one of us on the same ship. This is why work like reducing food waste moves me so deeply and why I'm so proud that together, we've helped raise funds to plant 70 million trees. The work continues! May we act as caretakers of this sacred creation. 🌎

Happy Earth Day 🌍 Today is a reminder of this extraordinary gift we've been given! This extraordinary Blue Planet we all call home, floating through the vast universe, carrying every ocean, every desert, every mountain, every forest. Every breath of life we have ever known. ✨ We were so moved by the historic new images of Earth from deep space that NASA shared from their Artemis II mission. The sheer MIRACLE of it all fills me with a deep, profound gratitude AND Serves as a powerful reminder that we each are called to be stewards of this precious miracle of life. The quiet, sacred web of life that breathes through us ALL. And let us also remember to be good stewards of EACH OTHER as we share this journey — every single one of us on the same ship. This is why work like reducing food waste moves me so deeply and why I'm so proud that together, we've helped raise funds to plant 70 million trees. The work continues! May we act as caretakers of this sacred creation. 🌎

Tony Robbins

12,734 просмотров • 4 месяцев назад

Anthropic ex-engineer runs an internal graph that costs $6 a month and catches what a $300,000 eval suite misses. No retrieval layer. Seven nodes and one rule about who's allowed to change their mind. He published the whole schema. His version starts from the opposite idea. A graph is not an execution order. It's a memory of why. Seven nodes. Every edge carries the reason it exists: > INTENT - states what the task is for. Never how > DECOMPOSE - splits it into steps, each with a stated assumption > WORKER - executes one step. Sees nothing else > AUDIT - checks the output against the assumption, not the goal > DRIFT - compares the current step to INTENT and flags divergence > LEDGER - stores every decision with the assumption that justified it > ROOT - holds the graph, and when an assumption breaks, re-runs every step built on it Six nodes act. One node remembers why they acted. Every step carries the assumption that made it correct, so a false one only reruns what stood on it. That's the entire design. A pipeline that forgets its reasons has to redo all of it or trust all of it. He replayed a month of agent runs. 4,100 steps, 380 built on an assumption that was wrong by day three. The old pipeline shipped all 380 and linked none of them. Everyone else builds graphs where output moves forward and the reasoning evaporates. He built one where the reason travels with the result. The blast radius is the output nobody else produces. The article below is the full build - node prompts, the assumption format, the invalidation rule that finds every step downstream of a broken one. Save it. You'll want it open in the other

Anthropic ex-engineer runs an internal graph that costs $6 a month and catches what a $300,000 eval suite misses. No retrieval layer. Seven nodes and one rule about who's allowed to change their mind. He published the whole schema. His version starts from the opposite idea. A graph is not an execution order. It's a memory of why. Seven nodes. Every edge carries the reason it exists: > INTENT - states what the task is for. Never how > DECOMPOSE - splits it into steps, each with a stated assumption > WORKER - executes one step. Sees nothing else > AUDIT - checks the output against the assumption, not the goal > DRIFT - compares the current step to INTENT and flags divergence > LEDGER - stores every decision with the assumption that justified it > ROOT - holds the graph, and when an assumption breaks, re-runs every step built on it Six nodes act. One node remembers why they acted. Every step carries the assumption that made it correct, so a false one only reruns what stood on it. That's the entire design. A pipeline that forgets its reasons has to redo all of it or trust all of it. He replayed a month of agent runs. 4,100 steps, 380 built on an assumption that was wrong by day three. The old pipeline shipped all 380 and linked none of them. Everyone else builds graphs where output moves forward and the reasoning evaporates. He built one where the reason travels with the result. The blast radius is the output nobody else produces. The article below is the full build - node prompts, the assumption format, the invalidation rule that finds every step downstream of a broken one. Save it. You'll want it open in the other

wast3

305,732 просмотров • 20 дней назад