正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

When you prompt an LLM for code, you get one deterministic program. However, the LLM actually defines a distribution over many programs, and existing methods discard it‼️ PPoT uses this distribution to extract free performance and efficiency gains. 🧵👇

Daniel Israel

1,385 subscribers

12,002 次观看 • 4 个月前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

5 条评论

Daniel Israel 的头像

Daniel Israel4 个月前

What is probabilistic programming? 🤔 It is just like regular coding, but instead of constants, it uses random variables with probability distributions. As a result, a single probabilistic program encodes a distribution over exponentially many deterministic programs.

Daniel Israel 的头像

Daniel Israel4 个月前

Normally, code generation frameworks prompt an LLM, test the output, and if the code fails, just resample from the LLM. PPoT can obtain multiple programs from just a single LLM sample by converting key tokens into random variables parameterized by the LLM's logits.

Daniel Israel 的头像

Daniel Israel4 个月前

PPoT leads to significant accuracy gains in a Best-of-N setting. Extracting additional PPoT samples per autoregressive sequence drastically improves the accuracy curve with zero GPU overhead.

Daniel Israel 的头像

Daniel Israel4 个月前

PPoT fundamentally alters inference scaling laws. Modeling the error rate with a power law reveals that PPoT exhibits a strictly better exponent in the number of LLM samples. The error rate drops faster, meaning the computational surplus from PPoT actually accelerates with additional test-time compute.

Daniel Israel 的头像

Daniel Israel4 个月前

It was a pleasure to collaborate with @PoorvaGarg11, @renatogeh, Todd Millstein, Kyle Richardson, and @guyvdb 📰 Paper : 🖥️ Code:

相关视频

How to enable full observability and automatic analytics for your LLM-based application. It takes one library + one line of code, and you get a ton of information for free. This is a no-brainer.

How to enable full observability and automatic analytics for your LLM-based application. It takes one library + one line of code, and you get a ton of information for free. This is a no-brainer.

Santiago

17,532 次观看 • 4 个月前

Free Mode is live in Higgsfield Supercomputer. The LLM inside is now free to use for brainstorming, prompt writing, scripting, research, and project planning. Credits kick in only when you generate. A Supercomputer-level LLM, yours to use.

Free Mode is live in Higgsfield Supercomputer. The LLM inside is now free to use for brainstorming, prompt writing, scripting, research, and project planning. Credits kick in only when you generate. A Supercomputer-level LLM, yours to use.

Higgsfield AI 🧩

266,061 次观看 • 2 个月前

He is 17 sold an LLM system to Anthropic for $2.2M - and at Stanford shared how to build and sell one from scratch: 00:20 - how to build an LLM system that Anthropic will buy for millions 08:34 - one prompt in Opus 5 and the LLM system is ready to sell in 60 minutes 23:47 - how to sell an LLM to Anthropic at 17 with no experience and reach $2.2M a year after watching I built my first LLM system in 60 minutes - and sold it to Anthropic for $120k. save & watch - the article below is one file that builds an LLM system and sells it to Anthropic for $2.2M.

He is 17 sold an LLM system to Anthropic for $2.2M - and at Stanford shared how to build and sell one from scratch: 00:20 - how to build an LLM system that Anthropic will buy for millions 08:34 - one prompt in Opus 5 and the LLM system is ready to sell in 60 minutes 23:47 - how to sell an LLM to Anthropic at 17 with no experience and reach $2.2M a year after watching I built my first LLM system in 60 minutes - and sold it to Anthropic for $120k. save & watch - the article below is one file that builds an LLM system and sells it to Anthropic for $2.2M.

Noisy

24,825 次观看 • 5 天前

He is 18 built an LLM and sold it to Anthropic for $3.2M - and at Stanford shared how to repeat it from scratch: 00:30 - how to build an LLM worth $100k in one evening 21:34 - how one LLM controls 1000 processes at the same time 43:47 - from first prompt to $3.2M from Anthropic after watching I built my first LLM in 60 minutes- a week later Anthropic offered me $100k for the system. save & watch - the article below shows how to build an LLM and sell it to Anthropic for $3.2M from scratch

He is 18 built an LLM and sold it to Anthropic for $3.2M - and at Stanford shared how to repeat it from scratch: 00:30 - how to build an LLM worth $100k in one evening 21:34 - how one LLM controls 1000 processes at the same time 43:47 - from first prompt to $3.2M from Anthropic after watching I built my first LLM in 60 minutes- a week later Anthropic offered me $100k for the system. save & watch - the article below shows how to build an LLM and sell it to Anthropic for $3.2M from scratch

Noisy

45,280 次观看 • 3 天前

🤔 Why do we still rely on the final layer of an LLM, when different layers encode different information? 🤔 In our new work, “Improving LLM Final Representations with Inter-Layer Geometry” (ICLR 2026 Workshop on Geometry-grounded Representation Learning and Generative Modeling) we show that actually, LLMs do not have one “best” layer. We introduce the Cayley-Encoder: an efficient and effective geometric encoder that learns one strong representation from all layer representations of the LLM, without biasing the representation toward any specific layer. While adding at most 0.1% learned parameters to the LLM, the Cayley-Encoder achieves large empirical gains over LoRA fine-tuning, final-layer representations, expensive attention-based aggregation, and methods that optimize specific layers for the task.

🤔 Why do we still rely on the final layer of an LLM, when different layers encode different information? 🤔 In our new work, “Improving LLM Final Representations with Inter-Layer Geometry” (ICLR 2026 Workshop on Geometry-grounded Representation Learning and Generative Modeling) we show that actually, LLMs do not have one “best” layer. We introduce the Cayley-Encoder: an efficient and effective geometric encoder that learns one strong representation from all layer representations of the LLM, without biasing the representation toward any specific layer. While adding at most 0.1% learned parameters to the LLM, the Cayley-Encoder achieves large empirical gains over LoRA fine-tuning, final-layer representations, expensive attention-based aggregation, and methods that optimize specific layers for the task.

Maya Bechler-Speicher

16,703 次观看 • 3 个月前

Learn to carry out red teaming attacks against your own LLM-based applications to spot and patch vulnerabilities! In our new short course, Red Teaming LLM Applications, Matteo Dora & Luca Martial of LLM testing company Giskard teach how to simulate malicious actions to discover vulnerabilities, and improve security. We start with prompt injection, where you can trick an LLM into bypassing safeguards to reveal private information, or say something inappropriate. There is no one-size-fits-all approach to security, but this course will help you identify some scenarios to protect against. We believe having red teaming capabilities widely known will result in greater transparency and safer LLM-based systems. However, we ask you to use the skills you gain from this course ethically. Please sign up here:

Learn to carry out red teaming attacks against your own LLM-based applications to spot and patch vulnerabilities! In our new short course, Red Teaming LLM Applications, Matteo Dora & Luca Martial of LLM testing company Giskard teach how to simulate malicious actions to discover vulnerabilities, and improve security. We start with prompt injection, where you can trick an LLM into bypassing safeguards to reveal private information, or say something inappropriate. There is no one-size-fits-all approach to security, but this course will help you identify some scenarios to protect against. We believe having red teaming capabilities widely known will result in greater transparency and safer LLM-based systems. However, we ask you to use the skills you gain from this course ethically. Please sign up here:

Andrew Ng

109,794 次观看 • 2 年前

This is a trillion-dollar industry, and you can't solve it with an LLM: • Forecasting • Fraud detection • Churn prediction Large Language Models are fundamentally bad at solving these problems. When you feed structured data into an LLM, it doesn't see relationships, and it treats every number, date, and foreign key as a token. That's why you always get garbage back. An LLM thinks your database is a Wikipedia article. It doesn't understand its structure or its relationships. GPT-4 scores 63% on relational prediction tasks. That's the best it can do, and that's pretty much useless. You can't expect real-world business value to come from summarizing Wikipedia articles.

This is a trillion-dollar industry, and you can't solve it with an LLM: • Forecasting • Fraud detection • Churn prediction Large Language Models are fundamentally bad at solving these problems. When you feed structured data into an LLM, it doesn't see relationships, and it treats every number, date, and foreign key as a token. That's why you always get garbage back. An LLM thinks your database is a Wikipedia article. It doesn't understand its structure or its relationships. GPT-4 scores 63% on relational prediction tasks. That's the best it can do, and that's pretty much useless. You can't expect real-world business value to come from summarizing Wikipedia articles.

Santiago

94,801 次观看 • 5 个月前

New short course: Evaluating AI Agents! Evals are important for driving AI system improvements, and in this course you'll learn to systematically assess and improve an AI agent’s performance. This is built in partnership with Arize AI and taught by John Gilhuly, Head of Developer Relations, and , Director of Product. I've often found evals to be a critical tool in the agent development process - they can be the difference between picking the right thing to work on vs. wasting weeks of effort. Whether you’re building a shopping assistant, coding agent, or research assistant, having a structured evaluation process helps you refine its performance systematically, rather than relying on random trial and error. This course shows you how to structure your evals to assess the performance of each component of an agent and its end-to-end performance. For each component, you select the appropriate evaluators, test examples, and performance metrics. This helps you identify areas for improvement both during development and in production. (If you're familiar with error analysis in supervised learning, think of this as adapting those ideas to agentic workflows.) In this course, you'll build an AI agent, and add observability to visualize and debug its steps. You’ll learn about code-based evals, in which you write code explicitly to test a certain step, as well as LLM-as-a-Judge evals, in which you prompt an LLM to efficiently come up with ways to evaluate more open-ended outputs. In detail, you’ll: - Understand key differences between evaluating LLM-based systems and traditional software testing. - Add observability to an agent by collecting traces of the steps taken by the agent and visualizing them - Choose the appropriate evaluator - code-based, LLM-as-a-Judge, human-annotation based - for each component. - Compute a convergence score to evaluate if your agent can respond to a query in an efficient number of steps. - Run structured experiments to improve the agent’s performance by exploring changes to the prompt, LLM model, or the agent’s logic. - Understand how to deploy these evaluation techniques to monitor the agent’s performance in production. By the end of this course, you’ll know how to trace AI agents, systematically evaluate them, and improve their performance. Please sign up here:

New short course: Evaluating AI Agents! Evals are important for driving AI system improvements, and in this course you'll learn to systematically assess and improve an AI agent’s performance. This is built in partnership with Arize AI and taught by John Gilhuly, Head of Developer Relations, and , Director of Product. I've often found evals to be a critical tool in the agent development process - they can be the difference between picking the right thing to work on vs. wasting weeks of effort. Whether you’re building a shopping assistant, coding agent, or research assistant, having a structured evaluation process helps you refine its performance systematically, rather than relying on random trial and error. This course shows you how to structure your evals to assess the performance of each component of an agent and its end-to-end performance. For each component, you select the appropriate evaluators, test examples, and performance metrics. This helps you identify areas for improvement both during development and in production. (If you're familiar with error analysis in supervised learning, think of this as adapting those ideas to agentic workflows.) In this course, you'll build an AI agent, and add observability to visualize and debug its steps. You’ll learn about code-based evals, in which you write code explicitly to test a certain step, as well as LLM-as-a-Judge evals, in which you prompt an LLM to efficiently come up with ways to evaluate more open-ended outputs. In detail, you’ll: - Understand key differences between evaluating LLM-based systems and traditional software testing. - Add observability to an agent by collecting traces of the steps taken by the agent and visualizing them - Choose the appropriate evaluator - code-based, LLM-as-a-Judge, human-annotation based - for each component. - Compute a convergence score to evaluate if your agent can respond to a query in an efficient number of steps. - Run structured experiments to improve the agent’s performance by exploring changes to the prompt, LLM model, or the agent’s logic. - Understand how to deploy these evaluation techniques to monitor the agent’s performance in production. By the end of this course, you’ll know how to trace AI agents, systematically evaluate them, and improve their performance. Please sign up here:

Andrew Ng

126,587 次观看 • 1 年前

🚨 FRANCOIS CHOLLET SAYS THE CURRENT LLM STACK IS THE WRONG PATH TO AGI Interviewer: “When will we accomplish the first definition of AGI?” Chollet confirms we are already on that exact trajectory, pointing out that: "Current technology can fully automate at human level or beyond any domain where you have verifiable rewards, right? And code... code being the first one." He then brings up the problem with LLMs, stating that: "It's possible in principle to build something that looks a lot like AGI on top of the LLM stack... I do believe, however, this would be the wrong thing to do because it would be very inefficient." Looking ahead to what the final architecture will actually require, he explains that: "AI research will have to trend towards not just efficiency, but in fact optimality over time. And for this reason, future AI in a few decades... it's not going to be this harness on top of a reasoning model on top of a base LLM."

🚨 FRANCOIS CHOLLET SAYS THE CURRENT LLM STACK IS THE WRONG PATH TO AGI Interviewer: “When will we accomplish the first definition of AGI?” Chollet confirms we are already on that exact trajectory, pointing out that: "Current technology can fully automate at human level or beyond any domain where you have verifiable rewards, right? And code... code being the first one." He then brings up the problem with LLMs, stating that: "It's possible in principle to build something that looks a lot like AGI on top of the LLM stack... I do believe, however, this would be the wrong thing to do because it would be very inefficient." Looking ahead to what the final architecture will actually require, he explains that: "AI research will have to trend towards not just efficiency, but in fact optimality over time. And for this reason, future AI in a few decades... it's not going to be this harness on top of a reasoning model on top of a base LLM."

Chris

15,578 次观看 • 5 个月前

Claude 3.7 Sonnet is insane... just one prompt, and you get these animated cards in HTML, JS, and CSS. You can try it for free in VSCode using the CodeGPT extension. Sharing the prompt in this thread 👇

Claude 3.7 Sonnet is insane... just one prompt, and you get these animated cards in HTML, JS, and CSS. You can try it for free in VSCode using the CodeGPT extension. Sharing the prompt in this thread 👇

Daniel San

117,724 次观看 • 1 年前

Claude Code with Sonnet 4.5 is actually incredible I gave it a prompt for a super complex app, and it one shot the entire thing In this video I walk you through how to use Claude Code to build a prompt library app you can start using immediately (no coding experience required):

Claude Code with Sonnet 4.5 is actually incredible I gave it a prompt for a super complex app, and it one shot the entire thing In this video I walk you through how to use Claude Code to build a prompt library app you can start using immediately (no coding experience required):

Alex Finn

40,955 次观看 • 11 个月前

I'm reiterating - I think a better interface for LLMs on desktop is an interactive notebook, not chat. I'm trying to build a better intuition for PCA and strongly feel an interactive notebook where I'm chatting with LLM + also modifying code/plotting figures is a much better interaction. Bonus points if you make notebook cells branchable (so if you get an error, you can ask LLM to fix it without adding that rotten context to the main branch of learning). Please steal this idea. Make an AI-powered, interactive, branching notebook.

I'm reiterating - I think a better interface for LLMs on desktop is an interactive notebook, not chat. I'm trying to build a better intuition for PCA and strongly feel an interactive notebook where I'm chatting with LLM + also modifying code/plotting figures is a much better interaction. Bonus points if you make notebook cells branchable (so if you get an error, you can ask LLM to fix it without adding that rotten context to the main branch of learning). Please steal this idea. Make an AI-powered, interactive, branching notebook.

Paras Chopra

58,232 次观看 • 1 年前

i found the secret to writing copy that SELLS with AI... it's not about the model, it's not about the prompt, and it doesn't matter how good you are at writing it's about feeding AI with world-class examples i built a full copywriting swipe file you can use with any LLM: - over 1,000 examples of the best emails, ads and landing pages copy - my 3-step process to use this inside AI projects - my super-prompt to extract the sauce from any copy reply "SWIPE" + retweet and i'll send it for free (must be following so i can DM)

i found the secret to writing copy that SELLS with AI... it's not about the model, it's not about the prompt, and it doesn't matter how good you are at writing it's about feeding AI with world-class examples i built a full copywriting swipe file you can use with any LLM: - over 1,000 examples of the best emails, ads and landing pages copy - my 3-step process to use this inside AI projects - my super-prompt to extract the sauce from any copy reply "SWIPE" + retweet and i'll send it for free (must be following so i can DM)

Machina

32,639 次观看 • 1 年前

My friend applied to 200 tech jobs in two years. No PhD. No Stanford. Last month Anthropic offered him $750,000. I asked him how he broke in from zero. He sent me a course that was never supposed to get out. A 3-hour video to build a full LLM from scratch. A developer teaches you exactly how LLMs like ChatGPT and Claude are actually built. I watched it last night. Halfway through, I realized it's embarrassingly simple to break into an AI lab. Bookmark this and read the article below. • 00:00 - intro to LLMs • 05:43 - LLM transformer architecture • 40:24 - training the LLM • 1:30:27 - modernizing the LLM • 2:33:53 - scaling the LLM

My friend applied to 200 tech jobs in two years. No PhD. No Stanford. Last month Anthropic offered him $750,000. I asked him how he broke in from zero. He sent me a course that was never supposed to get out. A 3-hour video to build a full LLM from scratch. A developer teaches you exactly how LLMs like ChatGPT and Claude are actually built. I watched it last night. Halfway through, I realized it's embarrassingly simple to break into an AI lab. Bookmark this and read the article below. • 00:00 - intro to LLMs • 05:43 - LLM transformer architecture • 40:24 - training the LLM • 1:30:27 - modernizing the LLM • 2:33:53 - scaling the LLM

Roan

279,295 次观看 • 2 个月前