正在加载视频...

视频加载失败

返回首页

Don't waste 2 years learning how Claude and ChatGPT actually work. Stanford just dropped a 1-hour course on the exact pipeline behind them. 0:00 - policy gradient basics 23:02 - PPO for LLM training 55:28 - how models learn chain of thought 1:02:50 - the architecture, explained look at...

hanakoxbt's profile picture

Hanako

14,139 subscribers

21,564 次观看 • 17 天前 •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

Stanford just dropped a 1 hour course on how it teaches its AI engineers to train LLMs from scratch. This is the exact way LLMs like ChatGPT & Claude are trained. • 00:00 - LLM Policy Gradient basics • 23:24 - PPO for LLM training • 56:20 - how LLMs learn Chain of Thought • 1:03:50- LLM architecture explained Anthropic pays $750,000/year to engineers who understand this exact knowledge of LLMs. Bookmark this & give 1 hours today, no matter what. It'll be the most productive thing you do this week.
1:18:27

Stanford just dropped a 1 hour course on how it teaches its AI engineers to train LLMs from scratch. This is the exact way LLMs like ChatGPT & Claude are trained. • 00:00 - LLM Policy Gradient basics • 23:24 - PPO for LLM training • 56:20 - how LLMs learn Chain of Thought • 1:03:50- LLM architecture explained Anthropic pays $750,000/year to engineers who understand this exact knowledge of LLMs. Bookmark this & give 1 hours today, no matter what. It'll be the most productive thing you do this week.

Roan

37,436 次观看 • 25 天前

Anthropic pays $750,000+ a year for engineers who can build LLM architectures from scratch. This 2-hour Stanford lecture gives you the exact pipeline LLM engineers get paid $750K/year for. Data + architecture + scaling laws + post-training. Bookmark it & watch today. Then read article below.
1:44:23

Anthropic pays $750,000+ a year for engineers who can build LLM architectures from scratch. This 2-hour Stanford lecture gives you the exact pipeline LLM engineers get paid $750K/year for. Data + architecture + scaling laws + post-training. Bookmark it & watch today. Then read article below.

Movez

562,688 次观看 • 3 个月前

Don't waste 2 years learning to build LLMs like Claude & ChatGPT. MIT dropped a 3 hour course on building LLMs from scratch for free. Training → Architecture → Scale. It covers everything about LLMs. • 00:00 - training an LLM • 1:20:46 - LLM architecture from scratch • 2:34:56 - scaling laws for LLMs Anthropic pays $750,000/year to AI engineers in 2026 who master this exact knowledge. Bookmark this & skip Netflix. Give it 4 hours today. Then read the article below.
3:13:05

Don't waste 2 years learning to build LLMs like Claude & ChatGPT. MIT dropped a 3 hour course on building LLMs from scratch for free. Training → Architecture → Scale. It covers everything about LLMs. • 00:00 - training an LLM • 1:20:46 - LLM architecture from scratch • 2:34:56 - scaling laws for LLMs Anthropic pays $750,000/year to AI engineers in 2026 who master this exact knowledge. Bookmark this & skip Netflix. Give it 4 hours today. Then read the article below.

Roan

85,383 次观看 • 18 天前

Don't waste 2 years learning to build LLMs like Claude & ChatGPT. Stanford just dropped a 2 hour course on how to build LLMs from scratch. 00:00 - break text into tokens 25:44 - decode a user prompt 35:40 - train the model end to end 1:16:47 - build the architecture from scratch Anthropic pays $750,000/year to engineers who understand this exact knowledge of LLMs. Watch it, then go further with the step-by-step guide below on how to build a system that improves itself.
2:34:50

Don't waste 2 years learning to build LLMs like Claude & ChatGPT. Stanford just dropped a 2 hour course on how to build LLMs from scratch. 00:00 - break text into tokens 25:44 - decode a user prompt 35:40 - train the model end to end 1:16:47 - build the architecture from scratch Anthropic pays $750,000/year to engineers who understand this exact knowledge of LLMs. Watch it, then go further with the step-by-step guide below on how to build a system that improves itself.

Anatoli Kopadze

104,744 次观看 • 7 天前

Head of Claude Code: "85% of our engineers are running dozens or hundreds of agents. The way you do it is graph engineering." In 40 minutes he explains how a single engineer now does the work of a whole team, how far this has gone inside Anthropic and where it goes next. This is something you can't skip if you don't want to be left behind. Watch it, then read the full guide on graph engineering below.
35:35

Head of Claude Code: "85% of our engineers are running dozens or hundreds of agents. The way you do it is graph engineering." In 40 minutes he explains how a single engineer now does the work of a whole team, how far this has gone inside Anthropic and where it goes next. This is something you can't skip if you don't want to be left behind. Watch it, then read the full guide on graph engineering below.

Anatoli Kopadze

1,723,784 次观看 • 1 个月前

context engineering vs graph engineering. every few months the list gets a new word and everyone treats it as a replacement for the last one. these two are not on the same list. one decides what the model sees this turn, the other decides what exists at all. the cleanest way to tell them apart is to ask what a single unit of work looks like. > context engineering is the window the window opens empty, every single time. you assemble what goes in it. the prompt, the docs, the history, the tool results. the assembling is the work. the window only grows. it never shrinks on its own, so eventually something gets dropped. usually from the middle. usually without telling you. then the turn ends and the window is thrown away. not archived, thrown away. the next turn opens empty again and you re-explain what you already explained. good context engineering is knowing what to leave out, not what to pack in. the unit of work is one window. > graph engineering is the structure the same material arrives from the same sources. instead of packing it into a window, you pull entities out of it, resolve the duplicates into one node, and write typed edges between them. nothing here is stored as text you hope to find again. it is stored as a thing with a name and its connections to other things. when the turn ends, the graph is still there. the next turn does not start from zero. it starts by querying what already exists, and the query walks edges instead of guessing at similarity. good graph engineering is deciding what counts as the same thing twice. the unit of work is one relationship. > they are not alternatives the graph is what refills the window. context engineering decides what fits. graph engineering decides what there is to choose from. remove the graph and every session starts blind. remove the context work and the best structure in the world arrives as an unreadable dump. that also tells you which one broke. the answer drifted from what you actually said, or forgot something from this same session. that is the window. the answer is coherent but invents a connection that does not exist, or cannot join two facts it has clearly seen. that is the structure. people debug the prompt because the prompt is the easiest thing to edit. it keeps taking the blame for failures that live a layer down. save this - then read the full breakdown below
0:40

context engineering vs graph engineering. every few months the list gets a new word and everyone treats it as a replacement for the last one. these two are not on the same list. one decides what the model sees this turn, the other decides what exists at all. the cleanest way to tell them apart is to ask what a single unit of work looks like. > context engineering is the window the window opens empty, every single time. you assemble what goes in it. the prompt, the docs, the history, the tool results. the assembling is the work. the window only grows. it never shrinks on its own, so eventually something gets dropped. usually from the middle. usually without telling you. then the turn ends and the window is thrown away. not archived, thrown away. the next turn opens empty again and you re-explain what you already explained. good context engineering is knowing what to leave out, not what to pack in. the unit of work is one window. > graph engineering is the structure the same material arrives from the same sources. instead of packing it into a window, you pull entities out of it, resolve the duplicates into one node, and write typed edges between them. nothing here is stored as text you hope to find again. it is stored as a thing with a name and its connections to other things. when the turn ends, the graph is still there. the next turn does not start from zero. it starts by querying what already exists, and the query walks edges instead of guessing at similarity. good graph engineering is deciding what counts as the same thing twice. the unit of work is one relationship. > they are not alternatives the graph is what refills the window. context engineering decides what fits. graph engineering decides what there is to choose from. remove the graph and every session starts blind. remove the context work and the best structure in the world arrives as an unreadable dump. that also tells you which one broke. the answer drifted from what you actually said, or forgot something from this same session. that is the window. the answer is coherent but invents a connection that does not exist, or cannot join two facts it has clearly seen. that is the structure. people debug the prompt because the prompt is the easiest thing to edit. it keeps taking the blame for failures that live a layer down. save this - then read the full breakdown below

Hanako

19,160 次观看 • 1 个月前

one agent → a loop → a graph → a system that throttles itself Google just put the whole ladder in one free course. most people never leave the first rung. 2 hours 4 minutes, taught by two Google Cloud engineers. no signup, no paywall. 17:55 - build your first agent 39:55 - loop engineering: iterate, check, break 1:13:25 - graph engineering 1:35:27 - agents that throttle themselves 1:56:19 - the full graph for multi-agent systems the fourth one is the step nobody talks about. an agent that cannot throttle itself is not autonomous, it is just unattended. everyone builds one agent and calls it done. this is the version where the agents wire themselves into a graph and decide their own rate. and the thing that decides when one of them is allowed to stop is not in the course. that part you build yourself. bookmark and watch it today, then read the eval setup below
2:04:19

one agent → a loop → a graph → a system that throttles itself Google just put the whole ladder in one free course. most people never leave the first rung. 2 hours 4 minutes, taught by two Google Cloud engineers. no signup, no paywall. 17:55 - build your first agent 39:55 - loop engineering: iterate, check, break 1:13:25 - graph engineering 1:35:27 - agents that throttle themselves 1:56:19 - the full graph for multi-agent systems the fourth one is the step nobody talks about. an agent that cannot throttle itself is not autonomous, it is just unattended. everyone builds one agent and calls it done. this is the version where the agents wire themselves into a graph and decide their own rate. and the thing that decides when one of them is allowed to stop is not in the course. that part you build yourself. bookmark and watch it today, then read the eval setup below

Hanako

53,692 次观看 • 26 天前

Anthropic pays $750,000+ a year for engineers who can build LLMs from scratch. Not how to prompt them. Not how to fine-tune them. Not how to build RAG pipelines. But how to build them from scratch. This 2-hour Stanford lecture teaches you everything. Scaling laws. Data collection. Architecture design. Post-training alignment. Free. From Stanford. Watch first. Then read this. The lecture is the theory. And this article shows you how to actually build it (with code) ↓
1:44:23

Anthropic pays $750,000+ a year for engineers who can build LLMs from scratch. Not how to prompt them. Not how to fine-tune them. Not how to build RAG pipelines. But how to build them from scratch. This 2-hour Stanford lecture teaches you everything. Scaling laws. Data collection. Architecture design. Post-training alignment. Free. From Stanford. Watch first. Then read this. The lecture is the theory. And this article shows you how to actually build it (with code) ↓

Rahul

423,243 次观看 • 2 个月前

Head of Google AI: "75% of our engineers are running dozens or hundreds of agents. The way you do it is Graph Engineering." in 50-minutes, he shows how one engineer can do a whole team’s work, how far this has gone at Google, and what’s next. this is something you can't skip if you don't want to be left behind. watch it, then read the full guide on graph engineering below.
56:58

Head of Google AI: "75% of our engineers are running dozens or hundreds of agents. The way you do it is Graph Engineering." in 50-minutes, he shows how one engineer can do a whole team’s work, how far this has gone at Google, and what’s next. this is something you can't skip if you don't want to be left behind. watch it, then read the full guide on graph engineering below.

Codez

76,425 次观看 • 29 天前

your agent reviewing its own work is not a check. it is a second opinion from the same source. this is the most common gap in agent systems and it hides in plain sight, because the step exists. there is a review. it just cannot do the thing you think it does. here is the mechanism. the model produced an output from a context. you then ask the same model, holding the same context, whether that output is correct. it answers fluently, because that is what it does. and the answer is drawn from the same distribution that produced the thing being judged. same weights, same window, same blind spots. if the reason the output is wrong is something the model does not know, the review does not know it either. if the reason is something the context does not contain, the review has the same context. the failure mode and the detector share a cause. > why it feels like it works because most of the time the output is fine, and the review says fine. agreement is not evidence of detection. a reviewer that says pass on everything agrees with reality most of the time too. what you actually want to measure is what happens on the cases that are wrong. that is the only place a check earns its name, and it is exactly the place where a self-review is weakest. there is research on this. Huang and colleagues at DeepMind showed at ICLR 2024 that intrinsic self-correction, revising without external grounding, does not reliably help and often makes things worse. > what to actually do move the check outside the model. a test that runs, a schema that validates, a file that exists or does not, an exit code from something you did not write. these are not smarter than the model. they are just not correlated with it, and that is the entire value. when the judgement genuinely needs a model, at minimum use a different family. same family means shared blind spots, and frontier judges measurably inflate scores for outputs that look like their own. and split the work by kind. anything objectively checkable goes to code. only the genuinely semantic calls go to a judge, and those get a rubric written as one line. a review inside the loop tells you the model is confident. a check outside it tells you whether the work is done. save this - then read the eval setup below
0:50

your agent reviewing its own work is not a check. it is a second opinion from the same source. this is the most common gap in agent systems and it hides in plain sight, because the step exists. there is a review. it just cannot do the thing you think it does. here is the mechanism. the model produced an output from a context. you then ask the same model, holding the same context, whether that output is correct. it answers fluently, because that is what it does. and the answer is drawn from the same distribution that produced the thing being judged. same weights, same window, same blind spots. if the reason the output is wrong is something the model does not know, the review does not know it either. if the reason is something the context does not contain, the review has the same context. the failure mode and the detector share a cause. > why it feels like it works because most of the time the output is fine, and the review says fine. agreement is not evidence of detection. a reviewer that says pass on everything agrees with reality most of the time too. what you actually want to measure is what happens on the cases that are wrong. that is the only place a check earns its name, and it is exactly the place where a self-review is weakest. there is research on this. Huang and colleagues at DeepMind showed at ICLR 2024 that intrinsic self-correction, revising without external grounding, does not reliably help and often makes things worse. > what to actually do move the check outside the model. a test that runs, a schema that validates, a file that exists or does not, an exit code from something you did not write. these are not smarter than the model. they are just not correlated with it, and that is the entire value. when the judgement genuinely needs a model, at minimum use a different family. same family means shared blind spots, and frontier judges measurably inflate scores for outputs that look like their own. and split the work by kind. anything objectively checkable goes to code. only the genuinely semantic calls go to a judge, and those get a rubric written as one line. a review inside the loop tells you the model is confident. a check outside it tells you whether the work is done. save this - then read the eval setup below

Hanako

14,325 次观看 • 24 天前

you read it and pictured yourself approving instead of watching approving what five agents, each on its own machine, handing work down the line, and what reaches you at the end is the last one's account of the four accounts before it the honest part of that piece says you do not own the harness that is the exact layer where a check on any of this would have to live a hundred and twenty a seat, three hundred at the top, about six times what you pay now you are not buying the work you are buying a shorter story about the work, told by the only party that watched it happen
1:41

you read it and pictured yourself approving instead of watching approving what five agents, each on its own machine, handing work down the line, and what reaches you at the end is the last one's account of the four accounts before it the honest part of that piece says you do not own the harness that is the exact layer where a check on any of this would have to live a hundred and twenty a seat, three hundred at the top, about six times what you pay now you are not buying the work you are buying a shorter story about the work, told by the only party that watched it happen

Lagosaur

21,838 次观看 • 3 天前

Loops vs. Graphs, clearly explained! loops are great, but they have a ceiling: a loop makes one unit of work better. it cannot decide which units exist. so you end up with a very good agent running the wrong three steps, in the wrong order, one at a time. Graph engineering fixes this by moving the decision up a layer: what runs, what runs at the same time, and what never runs at all. you need both. here's how it works: a graph splits your system into two kinds of decision. ↳ inside a unit: the loop. produce, check, correct, repeat until green ↳ between units: the graph. split, fan out, merge, gate, send back Prompts → Context → Harness → Loops → Graphs you get parallel work, isolated contexts, and steps that stop running when nothing needs them. the trick is being selective about what becomes a node. only spend a model where judgment lives. merging, ranking, deduping and schema checks are edges, and edges are code. free, instant, and they cannot be argued out of a verdict. a graph where every edge is an agent pays rent on its own wiring. one thing to know before you scale it. a graph has two return paths, and almost everyone builds one. ↳ the correction edge is short. a gate rejects one unit back to the step that produced it, and it fixes the run you are in ↳ the learning edge is long. an accepted result goes back to the splitter as a constraint, and it fixes every run after skip the second and you get a graph that is fast and never gets smarter. next week it starts from the same place with the same blind spots. and a smaller one that eats whole nights: when a unit fails, return that unit, not the batch. send back four slices because one failed and you have just rewritten three correct ones. do it twice in a run and the run never converges. 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 ↓
0:41

Loops vs. Graphs, clearly explained! loops are great, but they have a ceiling: a loop makes one unit of work better. it cannot decide which units exist. so you end up with a very good agent running the wrong three steps, in the wrong order, one at a time. Graph engineering fixes this by moving the decision up a layer: what runs, what runs at the same time, and what never runs at all. you need both. here's how it works: a graph splits your system into two kinds of decision. ↳ inside a unit: the loop. produce, check, correct, repeat until green ↳ between units: the graph. split, fan out, merge, gate, send back Prompts → Context → Harness → Loops → Graphs you get parallel work, isolated contexts, and steps that stop running when nothing needs them. the trick is being selective about what becomes a node. only spend a model where judgment lives. merging, ranking, deduping and schema checks are edges, and edges are code. free, instant, and they cannot be argued out of a verdict. a graph where every edge is an agent pays rent on its own wiring. one thing to know before you scale it. a graph has two return paths, and almost everyone builds one. ↳ the correction edge is short. a gate rejects one unit back to the step that produced it, and it fixes the run you are in ↳ the learning edge is long. an accepted result goes back to the splitter as a constraint, and it fixes every run after skip the second and you get a graph that is fast and never gets smarter. next week it starts from the same place with the same blind spots. and a smaller one that eats whole nights: when a unit fails, return that unit, not the batch. send back four slices because one failed and you have just rewritten three correct ones. do it twice in a run and the run never converges. 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

72,028 次观看 • 9 天前

Google Brain founder, Andrew Ng: "One of the most important skills of the future is the ability to tell a computer exactly what you want so it can do it for you." In this talk, he explains why the biggest opportunity isn't just using AI. It's learning how to turn work into agentic workflows, break tasks into steps, give agents the right tools and context, add evals, and make the whole system reliable. That's the difference between prompting an AI and engineering an agent that actually works. Watch it today, then read the full guide in the article below.
26:58

Google Brain founder, Andrew Ng: "One of the most important skills of the future is the ability to tell a computer exactly what you want so it can do it for you." In this talk, he explains why the biggest opportunity isn't just using AI. It's learning how to turn work into agentic workflows, break tasks into steps, give agents the right tools and context, add evals, and make the whole system reliable. That's the difference between prompting an AI and engineering an agent that actually works. Watch it today, then read the full guide in the article below.

Rahul

14,431 次观看 • 21 天前

The final lecture of my course is an intro to character training! This is a topic that I've been quietly very invested in for ~18 months, as it: * Has potential for high real world impact * Clearly used extensively at frontier labs * Almost no empirical literature exists * More accessible on academic compute This lecture covers what character training is, reviews model specs, constitutions, the differences, the motivations in real world events, some example research papers I like, and open questions in how it relates to post-training/model use generally. Hopefully this brings more people into the field (and reach out if you have questions). It is one of the more research-y chapters in my book, but one that I felt needed the reference. There is still so little, educational content on the topic online. 0:00 Intro 6:22 Part 1: Fundamentals — character, constitutions, and model specs 19:21 Part 2: Character training in practice 23:23 Part 3: Character elicitation without gradient steps 28:03 Part 4: Open questions (and the end of the course) 32:27 The course, complete Thanks for watching. No need to like and subscribe now that the course is done, you definitely wouldn't! h/t to Sharan for leading the technical work I got to do in the space, and Zafir Stojanovski for investing a lot of attention at this book chapter.
32:55

The final lecture of my course is an intro to character training! This is a topic that I've been quietly very invested in for ~18 months, as it: * Has potential for high real world impact * Clearly used extensively at frontier labs * Almost no empirical literature exists * More accessible on academic compute This lecture covers what character training is, reviews model specs, constitutions, the differences, the motivations in real world events, some example research papers I like, and open questions in how it relates to post-training/model use generally. Hopefully this brings more people into the field (and reach out if you have questions). It is one of the more research-y chapters in my book, but one that I felt needed the reference. There is still so little, educational content on the topic online. 0:00 Intro 6:22 Part 1: Fundamentals — character, constitutions, and model specs 19:21 Part 2: Character training in practice 23:23 Part 3: Character elicitation without gradient steps 28:03 Part 4: Open questions (and the end of the course) 32:27 The course, complete Thanks for watching. No need to like and subscribe now that the course is done, you definitely wouldn't! h/t to Sharan for leading the technical work I got to do in the space, and Zafir Stojanovski for investing a lot of attention at this book chapter.

Nathan Lambert

56,783 次观看 • 23 天前

Anthropic engineer at Stanford: “80% of our engineers are using self-improving loops. Now everyone is building agentic graphs.” in 1-hour Stanford lecture, Anthropic engineer reveals how Claude thinks - and why Graphs are the future. This 1-hour watch will replace your $500 course on agentic engineering. Watch it today, then read how to become a Graph Engineer in the article below.
1:11:30

Anthropic engineer at Stanford: “80% of our engineers are using self-improving loops. Now everyone is building agentic graphs.” in 1-hour Stanford lecture, Anthropic engineer reveals how Claude thinks - and why Graphs are the future. This 1-hour watch will replace your $500 course on agentic engineering. Watch it today, then read how to become a Graph Engineer in the article below.

Codez

27,902 次观看 • 28 天前

ANTHROPIC JUST GAVE AWAY THE COURSE EVERYONE ELSE CHARGES FOR it is a free 1 hour class on Loop Engineering built around Fable 5 this is the cleanest breakdown of Claude Code and agentic loops out there right now people are selling paid tutorials that cover less than this hour does you watch it once and you understand how to wire Claude into loops that run real work on their own start with the course, then go deeper on the step by step of building the loops yourself if you want the setup that keeps you ahead of everyone still typing code by hand, i put it all in my article below
1:01:16

ANTHROPIC JUST GAVE AWAY THE COURSE EVERYONE ELSE CHARGES FOR it is a free 1 hour class on Loop Engineering built around Fable 5 this is the cleanest breakdown of Claude Code and agentic loops out there right now people are selling paid tutorials that cover less than this hour does you watch it once and you understand how to wire Claude into loops that run real work on their own start with the course, then go deeper on the step by step of building the loops yourself if you want the setup that keeps you ahead of everyone still typing code by hand, i put it all in my article below

Atlas

31,005 次观看 • 1 个月前

"How do people learn to use Claude Code like Boris?" Claude Code creator Boris Cherny (Boris Cherny ) was asked by Diana Hu ( Diana ) at Startup School 2026. Boris Cherny: "Maybe don’t listen to the LinkedIn influencers. The thing about the model is that everyone is looking for one weird trick to make it work, but that simply doesn’t exist. There is nothing like that. You have to approach the model empirically. Give it a task that is too hard. Give it the tools to verify its work, just as you would use those tools yourself if you were doing the task. See where it struggles, and then fix the problem with better prompting, a skill, or if the model is missing context, an MCP connection that allows it to retrieve the context it needs. That is basically it. It sounds very simple. I think people tend to overthink and overengineer it because, in many ways, that is how we had to build systems in the past." ---- From "Y Combinator" YouTube channel, (full video link in comment)
1:19

"How do people learn to use Claude Code like Boris?" Claude Code creator Boris Cherny (Boris Cherny ) was asked by Diana Hu ( Diana ) at Startup School 2026. Boris Cherny: "Maybe don’t listen to the LinkedIn influencers. The thing about the model is that everyone is looking for one weird trick to make it work, but that simply doesn’t exist. There is nothing like that. You have to approach the model empirically. Give it a task that is too hard. Give it the tools to verify its work, just as you would use those tools yourself if you were doing the task. See where it struggles, and then fix the problem with better prompting, a skill, or if the model is missing context, an MCP connection that allows it to retrieve the context it needs. That is basically it. It sounds very simple. I think people tend to overthink and overengineer it because, in many ways, that is how we had to build systems in the past." ---- From "Y Combinator" YouTube channel, (full video link in comment)

Rohan Paul

55,255 次观看 • 1 个月前

OpenAI engineer: "At OpenAI, 85% of our engineers are running hundreds of agents with Codex. The way you do it is graph engineering." In 30-minutes lecture, he explains how a single engineer can now do the work of an entire team, how far this has already gone inside OpenAI, and where it goes next. This is something you can't skip if you don't want to be left behind. Watch it, then read the full guide on graph engineering below.
24:33

OpenAI engineer: "At OpenAI, 85% of our engineers are running hundreds of agents with Codex. The way you do it is graph engineering." In 30-minutes lecture, he explains how a single engineer can now do the work of an entire team, how far this has already gone inside OpenAI, and where it goes next. This is something you can't skip if you don't want to be left behind. Watch it, then read the full guide on graph engineering below.

Movez

203,914 次观看 • 20 天前

Agents vs. Graphs, clearly explained! spawning more agents is great, but it has a ceiling nobody says out loud: five agents is a count. a graph is a shape. only one of them changes the answer. point five agents at the same pile with the same window and they converge. the first one writes a finding, the rest read it, and all five reports centre on the same thing. you paid five times for one opinion with four echoes. Graph engineering fixes this by moving the decision up a layer: not how many agents, but who is allowed to look at what. you need both. here's how it works: ↳ the count buys you throughput. five things happening instead of one ↳ the shape buys you coverage. five different things happening instead of the same one five times Prompts → Context → Harness → Agents → Graphs the node that does this is the splitter, and it decides more than any other node in the system. cut a repository by folder and four workers audit the same three files. cut it by blast radius and each one sees something the others cannot. the trick is being selective about what each lane is allowed to see. separate contexts are not a nice-to-have, they are the mechanism. if two agents are meant to produce different things, they must not share a window. if they are meant to produce the same thing, you did not need two agents. one thing to know before you scale it. a branch that throws does not reject the batch. it resolves to null, and that is the containment. which means your merge quietly receives a short list. ↳ filter the nulls before the merge, or one dead lane poisons the whole result ↳ never index a merge by position. eight good branches and one failure will shift everything by one, silently skip that and the run looks like it worked. the output is just missing a lane, and nothing errored. and the one that eats whole nights: multi-agent setups can use up to fifteen times the total tokens of a single chat, because every lane reloads its own core. you are trading total tokens for a clean main window. usually the right trade, always a choice. 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 ↓
0:49

Agents vs. Graphs, clearly explained! spawning more agents is great, but it has a ceiling nobody says out loud: five agents is a count. a graph is a shape. only one of them changes the answer. point five agents at the same pile with the same window and they converge. the first one writes a finding, the rest read it, and all five reports centre on the same thing. you paid five times for one opinion with four echoes. Graph engineering fixes this by moving the decision up a layer: not how many agents, but who is allowed to look at what. you need both. here's how it works: ↳ the count buys you throughput. five things happening instead of one ↳ the shape buys you coverage. five different things happening instead of the same one five times Prompts → Context → Harness → Agents → Graphs the node that does this is the splitter, and it decides more than any other node in the system. cut a repository by folder and four workers audit the same three files. cut it by blast radius and each one sees something the others cannot. the trick is being selective about what each lane is allowed to see. separate contexts are not a nice-to-have, they are the mechanism. if two agents are meant to produce different things, they must not share a window. if they are meant to produce the same thing, you did not need two agents. one thing to know before you scale it. a branch that throws does not reject the batch. it resolves to null, and that is the containment. which means your merge quietly receives a short list. ↳ filter the nulls before the merge, or one dead lane poisons the whole result ↳ never index a merge by position. eight good branches and one failure will shift everything by one, silently skip that and the run looks like it worked. the output is just missing a lane, and nothing errored. and the one that eats whole nights: multi-agent setups can use up to fifteen times the total tokens of a single chat, because every lane reloads its own core. you are trading total tokens for a clean main window. usually the right trade, always a choice. 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

95,369 次观看 • 6 天前

somebody just torched their entire career for us someone inside Anthropic wrote down exactly how they wire their AIs to work as a team instead of one at a time, every number, and let the whole thing walk out the door it's called Graph Engineering: you stop asking one AI to do the whole job and start running a crew. one plans, a few go and look, one checks the others for lies their own measurement: the crew came back with 96% of the quality at 46% of the price. same models, same question. they only changed who is allowed to talk to whom Anthropic raised $65,000,000,000 to figure this out. it costs you $0 run one tonight in a normal chat, 3 moves: 1. give it the job and make it write the plan before it touches anything 2. run each step in its own fresh chat so nothing bleeds into the next 3. in one clean chat at the end, make it check the work against sources and mark whatever it could not confirm save this, then open the piece below: how many workers a job actually needs, and where one more starts making it worse ↓
0:34

somebody just torched their entire career for us someone inside Anthropic wrote down exactly how they wire their AIs to work as a team instead of one at a time, every number, and let the whole thing walk out the door it's called Graph Engineering: you stop asking one AI to do the whole job and start running a crew. one plans, a few go and look, one checks the others for lies their own measurement: the crew came back with 96% of the quality at 46% of the price. same models, same question. they only changed who is allowed to talk to whom Anthropic raised $65,000,000,000 to figure this out. it costs you $0 run one tonight in a normal chat, 3 moves: 1. give it the job and make it write the plan before it touches anything 2. run each step in its own fresh chat so nothing bleeds into the next 3. in one clean chat at the end, make it check the work against sources and mark whatever it could not confirm save this, then open the piece below: how many workers a job actually needs, and where one more starts making it worse ↓

Argona

1,379,115 次观看 • 16 天前