Nathan Lambert announces the launch of Trillium Labs, a new nonprofit dedicated to frontier AI research and research infrastructure. He discusses the decline of open science, the dynamics of AI…
I’ve had one too many discussions on the cyber risks of open models — where someone assumes that either banning open models happens in a silo (and bad actors will somehow actually be stopped) or that…
本期内容 这期节目围绕一个隐藏在多个新闻背后的共同问题：我们正在把越来越多的决策权交给 AI 系统，而对它们的了解还远远跟不上。OpenAI DevDay 带来了 Dots 常驻代理和大幅降价的 GPT-6.1 Sol，Anthropic 同步更新了 Claude Sonnet 5.5，模型的成本和能力都在快速移动。但与此同时，Nathan Lambert 在国会讲开源权力格局，Cal…
本期内容 这一期从实验室说到战场，从国会说到浏览器，从操作系统说到写作。Claude 独立完成了一个生物学原创发现，Pentagon 承认在一场致命袭击中过度依赖 AI 判断，Nathan Lambert 把给国会的简报公开发布，荷兰政府在推一个脱离微软的操作系统项目，Mistral 和 Firefox 联手做本地私有 AI 助手，还有一篇博文让人停下来想：当 AI…
TypeSafe AI发布"系统一"决策模型Jev：不写作文只下判断，输出不是文字而是带置信度的概率数字，首个模型输入价仅$0.042/百万token且输出免费，底子是未发表的"校准决策强化学习"，被看作给AI的if语句做工业化；风险是退回更彻底的黑箱，Simon Willison提醒千万别拿它给人排序。Nathan…
本期内容 这期节目覆盖了从隐私追踪到神经科学、从模型定价到政策证词的六个话题。ChatGPT 的广告追踪机制被独立研究者挖出，提醒我们重新审视对这类工具的信任预设。与此同时，Anthropic 和开源社区都在以各自的方式重塑 AI 能力版图。斯坦福的大脑研究让我们意识到，连最基础的科学假设都可能需要推翻。Ethan Mollick…
This episode is with Jean-Stanislas “JS” Denain of Epoch AI, who leads their Insights Team and is one of the people I find myself debating the state and trajectory of AI with more and more. We’ve had…
I was recently invited to brief a group of Congressional members and staff on the state of open-weight models in the lens of U.S.-China competition. I’m sharing my prepared remarks as a state of the…
Heute dominieren zwei Achsen: Inference-Effizienz auf vorhandener Hardware (Qwen, Kleinmodelle, MTP) und die wachsende Reife von AI-Agents in produktiven Umgebungen — von LinkedIn bis Unity. Dazu:…
We’re in an era where a few organizations are using thousands of concurrent agents to improve their processes and output. These organizations happen to be just the frontier AI labs, in particular…
ModelTalk regulars Nathan Lambert, who writes Interconnects, and writer Jasmine Sun join to work through a week in which Dario asked the industry to slow down and Trump called the whole thing a…
As AI became more powerful, it was inevitable that a different, growing group would start to take AI safety more seriously – what we did not know ahead of time, is which set of views they latched…
Housekeeping: Paid subscribers to Interconnects now get a permanent 40% discount on my book when purchasing at Manning.com. Access the code at the Interconnects perks page. Many AI optimists tend to…
本期内容 AI 正在更深地嵌进工作的每一个层级，而这期节目用六个真实案例把这件事说清楚了。从 Anthropic 给 Claude 输出打隐形水印，到英伟达斥资百亿想买下开发者大本营 Hugging Face，再到黑客用 AI 在漏洞传言阶段就发动攻击，每一件事都在提醒我们：AI…
本期内容 AI 系统的能力在快速扩张，但真正值得关注的问题往往藏在细节里：谁在往 AI 的知识库里塞偏向性信息？一个代理任务失控能花掉多少钱？ChatGPT 搜索悄悄改变了信息来源的逻辑，你知道吗？这期六件事，从模型交接的成本优化，到开源模型背后的资金博弈，试图帮你看清把越来越多控制权交给 AI 系统的代价和前提。 本期要点 - Nvidia 用线性变换实现跨模型 KV cache…
本期内容 本期从一个藏有追踪器的旧书讲起，揭开亚马逊收购稀有书籍用于 AI 训练数据的调查报道。六个话题串联起同一个方向：AI 的基础设施层正在快速重塑，数据来源从线上延伸到物理世界，本地推理模型日趋成熟，成本与路由问题成为真实运营挑战，而我们对 AI 能力边界的理解，仍需要持续校准。 本期要点 - 404 Media 记者把追踪器藏进二手古籍卖出，发现它最终进入亚马逊 AI…
本期内容 同一天，Google、OpenAI、DeepSeek 三家同时出手，Flash 降价、推理提速、开源框架入场，模型发布的节奏已经快到让版本号失去意义。但这期我们不只聊发布本身，还往后退了几步：有人把 AI 代理养在 Mac mini 里跑了一年，踩出了一套权限管理的实战经验；有研究者写完了一本 AI 教材，然后开始怀疑 AI 训练的方向是不是在优化错了目标；Hacker News…
Today on UpNext AI: a practitioner’s view of where AI writing still falls short, a demanding benchmark for enterprise agents, and new evidence on the limits of automated code-vulnerability detection.…
Nathan Lambert 是 AI 后训练领域的知名研究者，Interconnects 博客的主理人，曾任 Allen Institute for AI（AI2）研究员，现于创业公司深耕强化学习与语言模型。他在这篇文章中复盘了自己在 AI 辅助下写书的全过程。 他的核心洞察，相信很多读者也深有体会： 过去一年，新模型层出不穷，代码能力、agent 任务表现越来越强，但写作与表达能力却停滞不前…
Hosts: Lenar Kess, Damra Vol. Reasoning models bill you for tokens they won't let you read. A paper published yesterday shows those tokens are recoverable by replaying a response into a weaker…
Nathan Lambert spent three years as post-training lead at Ai2, where he built the OLMo models, and he writes Interconnects, one of the most-read technical newsletters in AI. He left Ai2 in June and…
Hosts: Lenar Kess, Damra Vol. OpenAI's security team published its Black Hat timeline of the Hugging Face incident, and within hours the argument had split: is a technical debrief about agents moving…
Hosts: Lenar Kess, Damra Vol. Yesterday afternoon, inside about two hours, Demis Hassabis moved from CEO of Google DeepMind to Chair, Koray Kavukcuoglu took over the lab, and Jeff Dean announced his…
本期内容 今天五篇覆盖了从密码学到工作流、从零售 AI 战略到开源力量格局的完整切面。Anthropic 用 Claude 发现了人类密码学家几十年没找到的数学漏洞，这不只是一个研究成果，而是一个信号：AI 开始在原创智识领域做出贡献了。与此同时，OpenAI 内部的工程文化正在被 AI 代理重塑，而 Target 的 AI…
Exciting news! My book trying to share post-training knowledge with the world is done and shipping soon. Order on Manning or Amazon. Thanks for the support. It’s currently the #1 AI book on Amazon…
本期内容 这期节目聊的是 AI 已经真实发生的问题，不是预测。超过一半的受访企业 AI 代理已经出过安全事故，权限设计的缺口比大多数团队意识到的更大。与此同时，Anthropic 发布 Claude Sonnet 5，把旗舰级代理能力带到了中间档价格，成本结构正在重写。开源 AI 正面临政策层面的关键六个月，而你精心积累的提示词技能库，可能也到了该清一清的时候。 本期要点 - Claude…
本期内容 这一期的五个话题，表面上分散，底下有一条共同的线：AI 不只是在变聪明，而是在改变它和人的关系结构。开源模型面临政策窗口收窄的压力，自由职业任务的 AI 通过率正在快速上升，ChatGPT 开始直接帮你生成可交付的成果，Claude Code 的 token 消耗揭示了工具背后的真实成本，Ethan Mollick…
本期内容 开源模型的最大威胁不是技术落后，而是监管。AI 工具在你不知情的情况下传输了什么数据？AI 完成真实付费项目的能力，到底到了哪一步？今天五件事围绕同一个问题展开：当 AI 从工具变成基础设施，新的管理风险和边界在哪里，你现在应该做什么准备。 本期要点 - 研究者 Nathan Lambert 警告开源模型面临监管关闭风险，前例已经存在，依赖 Llama 或 Mistral…
What happened with Anthropic's Mythos and Fable models? Why did Anthropic's approach to AI safety trigger such strong backlash? What does this mean for the future of AI regulation? What role does…
Nathan Lambert of https://www.interconnects.ai/, Jasmine Sun of https://jasmi.news/, and guest Ethan Ding of https://ethanding.substack.com/ check in Learn more about your ad choices. Visit…
Joining me today is Nathan Lambert, author of Interconnects AI and a post-training lead at the Allen Institute for AI. Nathan recently returned from a major tour of China’s leading AI labs, where he…
Nathan Lambert of https://www.interconnects.ai/ and Jasmine Sun of https://jasmi.news/ catch up. Learn more about your ad choices. Visit megaphone.fm/adchoices
Open models are often discussed as if they’re competing head-to-head with frontier systems. Are they catching up? Falling behind? Are they “good enough” yet? Nathan Lambert doesn’t believe open…
Swyx joined SAIL! Thank you SAIL Media, Prof. Tom Yeh, 8Lee, Hamid Bagheri, c9n, and many others for tuning into SAIL Live #6 with Nathan Lambert and Sebastian Raschka, PhD. Sharing here for the LS…
Nathan Lambert and Sebastian Raschka are machine learning researchers, engineers, and educators. Nathan is the post-training lead at the Allen Institute for AI (Ai2) and the author of The RLHF Book.…
本期播客来自《Lex Fridman Podcast》与机器学习专家 Sebastian Raschka 和 Nathan Lambert 对 2025 至 2026 年 人工智能发展态势 的深度对话。内容核心聚焦于 DeepSeek 等中国开源模型对全球竞争格局的冲击，并详细对比了 OpenAI、Anthropic 和 Google 等巨头在代码能力与推理模型上的差异。探讨了 比例定律…
Jasmine Sun of https://jasmi.news/ and Nathan Lambert of https://www.interconnects.ai/ report for duty. Athena makes a brief guest appearance before dipping for pilates. Jordan's flower app:…
Zixuan Li is Director of Product and genAI Strategy at Z.ai (also known as Zhipu 智谱 AI). The release of their benchmark-topping flagship model, GLM 4.5, was akin to “another DeepSeek moment,” in the…
In this special release episode, Matt sits down with Nathan Lambert and Luca Soldaini from Ai2 (the Allen Institute for AI) to break down one of the biggest open-source AI drops of the year: OLMo 3.…
The source is a transcript from a YouTube video by Nathan Lambert summarizing his talk at the PieTorch conference, which provided a recap and reflection on open models in 2025 , particularly…
The source is a transcript from a YouTube video by Nathan Lambert summarizing his talk at the PieTorch conference, which provided a recap and reflection on open models in 2025 , particularly…
Jasmine Sun (https://jasmi.news/), Nathan Lambert (https://www.interconnects.ai/), and special guest Afra Wang (https://afraw.substack.com/). Our article on 'China makes AI girlfriends and America…
Jasmine Sun of https://jasmi.news/ and Nathan Lambert of https://www.interconnects.ai/ discuss a wide variety of topics of interest around AI and culture. Outtro Music:…
Keegan McBride, Senior Policy Advisor in Emerging Technology and Geopolitics at the Tony Blair Institute, and Nathan Lambert, a post-training lead at the Allen Institute for AI, join Alan Rozenshein,…
This is a free preview of a paid episode. To hear more, visit www.nonzero.org 0:28 Nathan’s work at AI2—and his p(doom) 5:00 What does “open source AI” mean? 11:01 How Nathan taught a Llama new…
We first had Nathan on to give us his RLHF deep dive when he was joining AI2, and now he’s back to help us catch up on the evolution to RLVR (Reinforcement Learning with Verifiable Rewards), first…
(02:57) - Linda Yaccarino Steps Down as CEO of X (06:01) - Grok Spreads Antisemitic Rants on X (16:00) - Tariff Deadline Pushed to August (17:54) - F1's Red Bull Fires Christian Horner (26:54) -…
Nathan Lambert (https://www.interconnects.ai/), Jasmine Sun (https://jasmi.news/) and I kick off a new series 'Overfit' to talk how AI is shaping the future of media, why Nathan just needs $100m to…
Just how weird will the AI-powered future be? To discuss, ChinaTalk interviewed Nathan Lambert, who writes the Interconnects newsletter and researches AI at the Allen Institute. We get into… Why…
The conversation offers a sweeping overview of global AI competition, new releases in open-source and proprietary models, and the infrastructure arms race shaping the next wave of AGI development.…
An exciting week for Tom, who tells Nate about his company Hortus AI, whose mission is to make AI accountable to local communities. We cover a lot of classic Retort themes as Tom makes a case for…
Is China taking the lead in artificial intelligence? What’s the next big breakthrough in AI research? In this 17-minute episode of Brilliant Minds’ Digest, we distill insights from Lex Fridman’s…
Dylan Patel is the founder of SemiAnalysis, a research & analysis company specializing in semiconductors, GPUs, CPUs, and AI hardware. Nathan Lambert is a research scientist at the Allen Institute…
Nathan Lambert is the author of the popular AI newsletter Interconnects. He is also a research scientist who leads post-training at the Allen Institute for Artificial Intelligence, a research…
Tom and Nate sit down for a classic discussion of the role of AI in the modern philosophy of science. Much of this discussion is based on Thomas Samuel Kuhn's influential book The Structure of…
Nathan Lambert of the excellent https://www.interconnects.ai/ newsletter and the Allen Institute joins the pod for a rundown of the biggest AI stories of this year and next. We also talk about what…
We're back! Tom and Nate catch up after the Thanksgiving holiday. Our main question was -- what were the biggest AI stories of the year? We touch on the core themes of the show: infrastructure, AI…
Hey folks, Alex here, and oof what a 🔥🔥🔥 show we had today! I got to use my new breaking news button 3 times this show! And not only that, some of you may know that one of the absolutely biggest…
In this episode of The Cognitive Revolution, we dive deep into frontier post-training techniques for large language models with Nathan Lambert from the Allen Institute for AI. Nathan discusses the…
Tom and Nate catch up on core themes of AI after a somewhat unintended summer break. We discuss the moral groundings and philosophy of what we're building, our travels, The Anxious Generation, AGI…
Tom and Nate revisit one of their old ideas -- AI through the lens of public health infrastructure, and especially alignment. Sorry about Tom's glitchy audio, I figured it out after the fact that he…
Reinforcement learning through human feedback (RLHF) has come a long way. In this episode, research scientist Nathan Lambert talks to Jon Krohn about the technique’s origins of the technique. He also…
Nathan Lambert of the Interconnects substack and Allen Institute joins for a roundup where we get into: What DC should understand about the Bay Area AI engineer psyche What GPT4o and Google's AI Dev…
Tom and Nate discuss two major OpenAI happenings in the last week. The popular one, the chat assistant, and what it reveals about OpenAI's worldview. We pair this with discussion of OpenAI's new…
Tom and Nate discuss the shifting power landscape in AI. They try to discern what is special about Silicon Valley's grasp on the ecosystem and what other types of power (e.g. those in New York and…
Tom and Nate catch up after a few weeks off the pod. We discuss what it means for the pace and size of open models to get bigger and bigger. In some ways, this disillusionment is a great way to zoom…
Tom and Nate sit down to discuss Claude 3 and some updates on what it means to be open. Not surprisingly, we get into debating some different views. We cover Dune 2's impact on AI and have a brief…
Should AI be more open or closed? What does it mean to be open, anyway? And can France overtake China in AI?? Today I'm running a crossover episode with the Retort AI, hosted by AI Ethicist Tom…
We're excited to bring you something special today! Our first cross over episode brings some fresh energy to the podcast. Tom and Nate are joined by Jordan Schneider of ChinaTalk (A popular…
In 2023 we did a few Fundamentals episodes covering Benchmarks 101, Datasets 101, FlashAttention, and Transformers Math, and it turns out those were some of your evergreen favorites! So we are…
The end of the year is upon us! Tom and Nate bring a reflective mood to the podcast along with some surprises that may be a delight. Here are some links for the loyal fans: * RAND + executive order…
This week, Tom and Nate discuss some of the core and intriguing dynamics of AI. We discuss the history of the rationality movement and where Harry Potter fan fiction fits in, if AI will ever not feel…
Tom and Nate discuss some of the public institutions that form the bedrock of society -- education and roads -- and how AI is poised to shake them up. Some related reading on Interconnects,…
Tom and Nate discuss some of the most dominant metaphors in machine learning these days -- alchemy and deep learning's roots, the Oppenheimer film and a modern "Manhattan Project for AI", and of…
As first discussed on our May Emergency pod and leaked 4 days ago, Llama (renamed from LLaMA) was upgraded to Llama 2 (pretraining on 2 trillion tokens with 2x the context length - bigger than any…
Nathan Lambert is a PhD Candidate at UC Berkeley. Featured References Learning Accurate Long-term Dynamics for Model-based Reinforcement Learning Nathan O. Lambert, Albert Wilcox, Howard Zhang,…