Nathan Lambert

Research Scientist at Allen AI

Research @allen_ai, reasoning, open models, RL(VR/HF)... Contact via email. Writes @interconnectsai, @readsail Wrote The RLHF Book, 🏔️🏃‍♂️

Seattle

Twitter Blog Substack

Posts tracked from 1/5/2026 to 3/30/2026

Posts Analyzed

Platforms

Total Followers

78,983

Positive Sentiment

48%

38 positive of 80

Tools Discussed by Nathan Lambert

Fig

dev-tools · 25,133 stars

NVIDIA

llm-provider

Claude Code

dev-tools

DeepSeek

llm-provider · 102,417 stars

Share: Twitter LinkedIn

Social Reach79.0K total

79.0K

followers

Developer Profile

npm packages

175.2M

weekly downloads

Most Engaging Post

twitter

GPT 5.4 is the first time I've used codex for multiple hours straight and not ragequit back to claude code.

The user expresses satisfaction with GPT 5.4, noting it allows for extended use of Codex without frustration.

1,6523/7/2026View

Posts by Platform3.3 posts/week

Latest open artifacts (#20): New orgs! New types of models! With Nemotron Super, Sarvam, Cohere Transcribe, & others

Similar to the first post, it discusses the introduction of new organizations and models in the AI field, mentioning Nemotron Super, Sarvam, and Cohere Transcribe.

3/30/2026new modelsai development

Lossy self-improvement

This post presents a case for the reality of self-improvement while asserting that it does not lead to a fast takeoff in AI.

3/22/2026self-improvementai safety

GPT 5.4 is a big step for Codex

The author evaluates GPT 5.4's advancements in Codex and discusses their preference for Claude.

3/18/2026codexgpt 5.4

What comes next with open models

The post discusses the current state and future prospects of open models in the context of language model industrialization.

3/16/2026open modelslanguage models

Dean Ball on open models and government control

The post discusses the implications of the Anthropic v. Department of War case on the future of open models and government control.

3/6/2026open modelsgovernment control

Avg engagement: 0Total: 0

Activity Timeline

Recent Posts

rss

Latest open artifacts (#20): New orgs! New types of models! With Nemotron Super, Sarvam, Cohere Transcribe, & others

3/30/2026new modelsai development

substack

Latest open artifacts (#20): New orgs! New types of models! With Nemotron Super, Sarvam, Cohere Transcribe, & others

3/30/2026new modelsai development

substack

Lossy self-improvement

3/22/2026self-improvementai safety

rss

Lossy self-improvement

3/22/2026self-improvementai safety

rss

GPT 5.4 is a big step for Codex

3/18/2026codexgpt 5.4

substack

GPT 5.4 is a big step for Codex

3/18/2026gpt 5.4codex

rss

What comes next with open models

3/16/2026open modelslanguage models

substack

What comes next with open models

3/16/2026open modelslanguage models

twitter

@eliebakouch congrats & thanks to all your wonderful contributions to the ecosystem! HuggingFace alumni club

17 3/9/2026ai communityhuggingface

twitter

@N8Programs @seconds_0 relatable, you're hired

8 3/9/2026relatability

twitter

@noah_vandal Im saying that the singularity isn't real and isn't what is happening, I think the technical definition is fine (near infinite rapid improvement w/o much human involvement)

7 3/9/2026singularitytechnical definition

twitter

People overuse the singularity as the term to describe the large acceleration in ai progress that'll come due to better agents. There are still real points of friction that the current models don't ad

90 3/9/2026ai progresssingularity

twitter

@vikhyatk disagree.

12 3/8/2026disagreement

twitter

I was using Claude Code for some csv manipulation & it errored and dumped the entire Opus 4.6 reasoning trace to me. I'm surprised how similar closed models' reasoning behaviors are to far inferior o

297 3/8/2026claude codecsv

twitter

@romainhuet I tried sub agent and it said something weird. For research aide, sub agent is essential. Maybe was a skill issue on my part.

15 3/8/2026sub agentresearch

twitter

@tszzl previous codex's failed too much on little things like git or other "agentic" tasks irrelevant to code quality

126 3/7/2026codextasks

twitter

GPT 5.4 is the first time I've used codex for multiple hours straight and not ragequit back to claude code.

1652 3/7/2026codexgpt

twitter

@saurabh_shah2 told you so

10 3/7/2026predictionsocial media

twitter

@poof_eth im doing a lot of things like rlhfbook maitenace or building markdown slides or deep analysis w subagents. I havent been doing a lot of SWE

3 3/7/2026rlhfsubagents

twitter

GPT 5.4 in codex cli/app is much more approachable than any of their models that came before. This is really big for them, excited to keep trying it vis a vis Claude as my agent daily driver.

250 3/7/2026gptcodex