Nathan Lambert on AI
16 quotes from Nathan Lambert on the wall, 2022 to 2026, newest first. Each one is verbatim and links to where it was said. Sources include Interconnects AI, interconnects.ai, Interconnects (Nathan Lambert).
- 2026
Good ideas can be much more valuable than good execution in software
From Nathan Lambert's essay "I expect rapid progress but not towards general superintelligence" on Interconnects, October 2026, discussing how coding agents are shifting the balance between research ideas and engineering execution in AI.
Interconnects (Nathan Lambert)I expect rapid progress but not towards general superintelligence
Lots of what they buy is frankly crap
From an Interconnects essay by Nathan Lambert on AI research acceleration, October 2026, describing the low quality of RL training data sold by data companies, which leading labs still see clear return on investment from buying.
Interconnects (Nathan Lambert)I expect rapid progress but not towards general superintelligence
- 2025
Abilities will develop more slowly than products
Nathan Lambert, writing on Interconnects in "GPT-5 and the arc of progress" about the GPT-5 release in August 2025, summarizing how AI product improvements are outpacing raw capability gains.
A world where the most available models are the hardest to trust
Nathan Lambert, in a 2025 Interconnects essay proposing 'The American DeepSeek Project,' warned about a future dominated by powerful closed American models alongside widespread open Chinese models.
For the time being, the default is Qwen
Nathan Lambert, researcher at Ai2, in his Interconnects newsletter on Alibaba's Qwen 3 release overtaking Llama as the leading open-weight model suite, April 2025.
Llama is no longer the open standard
Nathan Lambert, in his Interconnects newsletter analysis of Meta's Llama 4 release (April 2025), on Meta losing its position in open-weight AI.
Interconnects AILlama 4: Did Meta just push the panic button?
Now we really know what Ilya saw
Nathan Lambert, in Interconnects analysis of GPT-4.5's release, February 2025, referring to reports that Ilya Sutskever had foreseen limits to pretraining scaling.
- 2024
Scaling as we know it is ending
From Nathan Lambert's 2024 year-in-review on Interconnects, describing the impact of OpenAI's o1 model on the AI training paradigm.
Scaling works and that doesn't solve all the economic problems for AI
Nathan Lambert, AI researcher and writer at Interconnects, in his post "Scaling realities" on the gap between technical scaling progress and commercial payoff, November 2024.
Every time someone is talking about AGI they're talking about vibes rather than reality
Nathan Lambert, trained in reinforcement learning, in his Interconnects essay "AGI is what you want it to be", April 2024, questioning the meaning of AGI.
- 2023
The numbers are marketing, not science
Nathan Lambert, writing on Interconnects AI in December 2023, criticized Microsoft and Google's benchmark comparisons between GPT-4 and Gemini, made in the week after Google announced the Gemini model suite.
RLHF is the smiling face of the Shoggoth Monster
Nathan Lambert, writing in his Interconnects essay "Reckoning with the Shoggoth of AI" in November 2023, used the Shoggoth meme to describe how RLHF masks the underlying nature of large language models.
- 2022
Data is the moat for ML systems right now
Nathan Lambert, writing on Interconnects in his December 2022 essay "Predicting machine learning moats", on what protects machine learning companies from competitors.
They hold the keys to the kingdom
Nathan Lambert, working at HuggingFace, on his Interconnects blog, December 2022, describing Surge AI's dominance in providing human-feedback data for RLHF, which he calls becoming "the bank of RLHF".
interconnects.aiClosed-API vs Open-source continues: RLHF, ChatGPT, data moats
There'll be an open-source ChatGPT within 6 months, surely
Nathan Lambert of HuggingFace predicting the pace of open-source AI catching up to ChatGPT, in a December 2022 blog post on Interconnects. He added that user interfaces and data longevity engrain OpenAI's advantage.
interconnects.aiClosed-API vs Open-source continues: RLHF, ChatGPT, data moats
Mentioned in other quotes 5
Quotes by others that refer to Nathan Lambert
- 2025
They hire smart young people as soldiers and deliberately burn them out with extremely long hours
Ross Taylor, former Galactica lead at Meta now doing a startup, describing the internal culture of major AI labs, comparing them to investment banks, in topic notes for an Interconnects podcast interview with Nathan Lambert, July 2025.
- 2024
Safety is not a model property
Dean W. Ball, AI policy researcher, discussing California's SB 1047 bill and its rules on model developers in an interview with Nathan Lambert on Interconnects, June 2024.
- 2023
Line from a leaked internal Google memo, quoted and discussed by Nathan Lambert on Interconnects in 2023, arguing Google lacks a unique technical advantage in large language models.
More similar to training a dog than to ordinary programming
From OpenAI's blog post clarifying ChatGPT's current and future behaviors, quoted in Nathan Lambert's 2023 Interconnects essay on AI alignment discourse, February 2023.
interconnects.ai"AI alignment" and uncalibrated discourse on AI
Corporations don't get to decide if we have competitors as a species
Quoted by Nathan Lambert in a 2023 essay on Interconnects discussing AI alignment and safety discourse.
interconnects.ai"AI alignment" and uncalibrated discourse on AI