korrents
Nathan Lambert

What Nathan Lambert thinks about LLMs

@nathan-lambert · 55 positions · 1 change of mind · On ppll ↗

Machine-learning researcher on open language models; writes the Interconnects newsletter and the RLHF Book, after leading post-training at Ai2.

Nathan Lambert did not write this page.

We collected these quotes from things they published elsewhere, and every quote links to where it was said. They have no account here and have not endorsed this site. Quotes are word for word; the short line under each one is our own restatement, not their wording. Their own site. Is this you? Claim it or ask us to remove it. Or tell us what is wrong here.

Filter LLMs · Newest first
Topic
All 55LLMs 30open source 11AI alignment 6America 6China 5reinforcement learning 5OpenAI 4AGI 3Anthropic 3
23 more topics
innovation 3AI and science 2infrastructure 2recursive self-improvement 2scaling laws 2semiconductors 2AI agents 1AI and jobs 1AI writing 1censorship 1creativity 1cybersecurity 1education 1energy 1measuring intelligence 1motivation 1neural networks 1politics 1quantum mechanics 1Silicon Valley 1startups 1taste 1writing 1
Sort
NewestOldest

Only Nathan Lambert on LLMs. See everyone on LLMs

30 dated positions, 2024 to 2026, in their own words. Our reading of what Nathan Lambert has said — not written or endorsed by them.

  1. Our readingThe AI research cycle from idea to experiment will likely soon be 10x faster, but that may not make researchers 10x more productive in the field's net understanding.

    Their wordsAnd my kind of statement is that I think the cycle from of having an idea and doing the experimentation to test it, I agree will be 10X faster very soon. But I don’t necessarily agree that I would say that AI researchers will be 10X more productive in net, which I would describe as the pace of the field’s complete understanding. And understanding is a different axis from just continuing to scale models. I think that’s one of my core confusions on the AI research side.

    ↗Debating RSI, the US-China Gap, and Jaggedness with JS Denain of Epoch AIinterconnects.ai

  2. 1 day earlier

  3. Our readingChinese AI companies are the clear leaders in open-weight models.

    Their wordsChinese AI companies are the clear leaders in open weight models.

    ↗The current balance of power in open modelsinterconnects.ai

  5. Our readingThe capabilities gap between open models and the closed models available to users has been decreasing over the last three years.

    Their wordsThe capabilities gap from open to closed models available to users has been decreasing over the last 3 years.

    ↗The current balance of power in open modelsinterconnects.ai

  6. 2 days earlier

  7. Our readingRecursive self-improvement in AI will mainly make existing model capability cheaper rather than raise peak intelligence.

    recursive self-improvement

    Their wordsRSI is poised to make modern LLMs vastly cheaper. Trends that have shown LLMs get exponentially cheaper at a given intelligence are likely to accelerate.

    ↗Why I still haven’t bought into true RSIinterconnects.ai 2nd of 3 in this piece

  8. 8 days earlier

  9. Our readingThe performance gap between open and closed AI models has narrowed to roughly 4-6 months.

    Their wordsThe open-closed model gap has reduced in recent years, and is now at roughly 4-6 months.

    ↗Open-Source AI & Open Models Reading Listinterconnects.ai 1st of 3 in this piece

  11. Our readingSince 2024, the leading open-weight AI models have come from Chinese labs rather than American ones.

    America

    Their wordsThe leading open models have all come from Chinese labs since ~2024.

    ↗Open-Source AI & Open Models Reading Listinterconnects.ai 2nd of 3 in this piece

  13. Our readingDistillation is the most significant debate surrounding open AI models in 2026.

    Their wordsDistillation - the process of training on output tokens from another model - is the single most eventful debate around open models in 2026.

    ↗Open-Source AI & Open Models Reading Listinterconnects.ai 3rd of 3 in this piece

  14. 1 day earlier

  15. Our readingAI progress is jagged: models are becoming superhuman at math and software engineering while remaining far weaker at intuition, creativity, and other forms of human reasoning.

    creativity

    Their wordsAI has always been very jagged, and we are making models which are superhuman goal-seekers at math and software engineering, but they have massive limitations on intuitions, creativity, and other types of reasoning that humans are strong at.

    ↗One resignation turned the embers of AI fear into a wildfireinterconnects.ai 3rd of 3 in this piece

  16. 2 days earlier

  17. Our readingFrontier Chinese AI labs are moving toward more restrictive licensing terms for their open models.

    Their wordsChinese model makers at the frontier, however, are becoming more restrictive

    ↗Latest open artifacts (#24): Motif-3, GLM-5.3, Hy4-preview and open model licensesinterconnects.ai 1st of 2 in this piece

  19. Our readingHybrid and sparse-attention model architectures will become more widely adopted as the ecosystem catches up.

    Their wordswe expect similar architectures to become more popular and the ecosystem to fix integrations by the time Qwen4 drops.

    ↗Latest open artifacts (#24): Motif-3, GLM-5.3, Hy4-preview and open model licensesinterconnects.ai 2nd of 2 in this piece

  20. 3 weeks earlier

  21. Our readingOpen models will end up serving a long-tail of specialized uses while closed models retain the most valuable applications.

    Their wordsopen models are still incredibly useful, but fill a long-tail ecosystem relative to the closed counterparts that have monopoly ownership stakes in the most valuable areas like knowledge work collaboration, drug discovery, SWE, etc.

    ↗Teaching Everyone to Fish for Tokensinterconnects.ai 2nd of 3 in this piece

  23. Our readingReleasing free access to AI intelligence is one of the strongest available business strategies.

    Their wordsreleasing access to intelligence is one of the strongest business strategies available.

    ↗Teaching Everyone to Fish for Tokensinterconnects.ai 3rd of 3 in this piece

  24. 3 days earlier

  25. Our readingOpenAI and Anthropic's internal, unreleased models are likely far more capable than Z.ai and Moonshot AI's public models.

    AnthropicOpenAI

    Their wordsIt is very, very likely that OpenAI and Anthropic have far better internal models than Z.ai and Moonshot AI.

    ↗GLM-5.3: How Chinese labs keep stride with the frontierinterconnects.ai 1st of 3 in this piece

  27. Our readingStaged, controlled rollout of a powerful AI model provides little safety benefit once open-weight versions of similar capability are inevitably released.

    Their wordsAt the end of the day, this type of safety barely matters when true open-weights are coming.

    ↗GLM-5.3: How Chinese labs keep stride with the frontierinterconnects.ai 3rd of 3 in this piece

  28. 2 days earlier

  29. Our readingProgress toward AI autonomously solving major open scientific problems requires first mastering the organization and presentation of established knowledge in long-form writing.

    Their wordsModels being stagnant in long-form, non-fiction writing should be alarming to those reliant on models autonomously solving grand, open science problems in the near future.

    ↗I wrote an AI textbook — how long until AI can do it better?interconnects.ai 1st of 3 in this piece

  30. 18 months earlier

  31. Our readingData processing, filtering and quality, not architecture, is the number one determinant of how good a language model turns out.

    Their wordsAnd we'll get into the details of the models and again and again as we try to get deeper into how the models were trained, we will say things like the data processing, data filtering data quality is the number one determinant of the model quality.

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 1st of 44 in this recording

  33. Our readingA published training-run cost always understates the truth, because any notable model takes two to four times that run's compute in experiments alone.

    Their wordsAccepted practice is that for any given model that is a notable advancement, you're going to do two to 4x compute of the full training run in experiments alone.

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 5th of 44 in this recording

  35. Our readingExport controls cannot stop China from training frontier models; what they can do is cap how much AI China gets to run.

    Chinaquantum mechanics

    Their wordsThere's not many worlds where China cannot train AI models. I think export controls are decapping the amount of compute or the density of compute that China can have.

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 6th of 44 in this recording

  37. Our readingLanguage models are already a form of AGI; what the labs are chasing is a further step, not the arrival of general intelligence.

    measuring intelligenceAGI

    Their wordsI think my personal definition of AGI is much simpler. I think language models are a form of AGI and all of this super powerful stuff is a next step that's great if we get these tools. But a language model has so much value in so many domains that it's a general intelligence to me.

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 8th of 44 in this recording

  39. Our readingLanguage models have not made misinformation meaningfully worse, because distribution, not the cost of writing convincing text, was always the limiting factor.

    Their wordsThere's some research that shows that the distribution is actually the limiting factor. So language models haven't yet made misinformation particularly change the equation there.

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 9th of 44 in this recording

  41. Our readingOpen frontier models will keep being released whether or not anyone wants to stop them, and stopping them would only leave the world less prepared.

    Their wordsthese open models are probably going to keep coming for the time being, whether or not we want to stop them, and stopping them might make it even worse and harder to prepare.

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 19th of 44 in this recording

  43. Our readingCensoring a fact out of a language model is practically impossible, because you would first have to remove that fact from the internet.

    censorship

    Their wordsI almost think it's practically impossible because you effectively have to remove them from the internet.

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 21st of 44 in this recording

  45. Our readingDistilling from another lab's model is standard industry practice, not cheating; who you distill from just depends on whether you have products to protect.

    Their wordsDistillation is standard practice in industry. Whether or not, if you're at a closed lab where you care about terms of service and IP closely, you distill from your own models.

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 29th of 44 in this recording

  47. Our readingOpenAI's claim that DeepSeek trained on its outputs is narrative management: other startups bootstrapped exactly that way and were never banned for it.

    OpenAIstartups

    Their wordsI think that they're trying to shift the narrative. They're trying to protect themselves. We saw this years ago when ByteDance was actually banned from some OpenAI APIs for training on outputs. There's other AI startups that most people, if you're in the AI culture, were like they just told us they trained on OpenAI outputs and they never got banned.

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 30th of 44 in this recording

  49. Our readingThe biggest untapped business in AI is advertising inside model outputs, and nobody has yet worked out technically how to place an ad there.

    Their wordsThe short-term that company that could make the most money is the one that figures out what advertising targeting method works for language model generations.

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 36th of 44 in this recording

  51. Our readingReasoning will generalize the way instruction tuning did: add enough verifiable domains and, at some point nobody can yet locate, the rest start working on their own.

    Their wordsAnd the history of NLP and language processing instruction, tuning and tasks per language model used to be like one language model did one task, and then in the instruction tuning literature, there's this point where you start adding more and more tasks together where it just starts to generalize to every task. And we don't know where on this curve we are.

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 37th of 44 in this recording

  53. Our readingHumans keep their place in AI as judges rather than authors, because telling which of two answers is better is far easier than writing a good one.

    reinforcement learning

    Their wordsAnd humans are actually very good at reading or judging between two things versus... This goes back to the core of what RLHF and preference tuning is that it's hard to generate a good answer for a lot of problems, but it's easy to see which one is better.

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 41st of 44 in this recording

  55. Our readingOpen-source AI is still an ideological mission rather than an ecosystem, because it has none of the feedback loops that make open-source software compound.

    open sourceAmericamotivation

    Their wordsuntil there are feedback loops of open source AI, it seems like mostly an ideological mission. People like Mark Zuckerberg, which is like America needs this and I agree with him, but in the time where the motivation ideologically is high, we need to capitalize and build this ecosystem around, what benefits do you get from seeing the language model data?

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 42nd of 44 in this recording

  56. 2 months earlier

  57. Our reading · no longer heldOpenAI's o1 does an online search over reasoning steps at test time, unlike a traditional autoregressive language model.

    OpenAIreinforcement learning

    No longer holdsThis is the argument that all of the “search” in OpenAI’s o1 (and some of the replications) is all within the reinforcement learning training behavior.

    SaidIn my first post on o1, I tried to take OpenAI’s corporate communications and the media of their employees at face value to figure out what is going on: What would it mean for OpenAI’s o1 to be using search?

    ↗OpenAI's o1 using "search" was a PSYOPinterconnects.ai

  58. 3 months earlier

  59. Our readingOpenAI's o1 does an online search over reasoning steps at test time, unlike a traditional autoregressive language model.

    OpenAI

    Their wordsUnlike traditional autoregressive language models, it is doing an online search for the user.

    ↗Reverse engineering OpenAI’s o1interconnects.ai