AI researchers to follow on X, and where to check what they post

The AI researchers to follow on X for research, rather than hype, include Andrej Karpathy, Lilian Weng, Noam Brown, Jan Leike, François Chollet and AK, who posts new papers as they appear. Pair them with a news source that checks the papers behind the posts: ClaudeGPT Si is a free publication that explains new AI research in plain English and links each paper.

Why both? A researcher's post is a fast signal. It is also one person's take, often written in a few lines. The explainer is where you find out what the paper actually tested.

The list

Every account below comes from the "AI labs and researchers" group in TechTwitter IO's guide to who to follow. Descriptions are based on that guide.

Researchers who explain

Andrej Karpathy was a founding member of OpenAI and later led AI at Tesla. He coined "vibe coding", and his long posts are some of the clearest explanations of how models work. Lilian Weng, a researcher at OpenAI, writes Lil'Log, a blog that goes deep on machine learning topics. Jeremy Howard (@jeremyphoward), co-founder of fast.ai and Answer.AI, teaches deep learning for free.

Researchers at the frontier labs

Noam Brown works on reasoning at OpenAI and co-created the o-series models, as well as the AI that beat top poker pros. Jan Leike is at Anthropic, after earlier roles at OpenAI and DeepMind. Demis Hassabis (@demishassabis) co-founded Google DeepMind and shared the 2024 Nobel Prize in Chemistry.

The paper spotters

AK posts the day's new AI papers, which makes the account the quickest way to see research as it lands; AK also curates Hugging Face Papers with the community. Nathan Lambert co-led Olmo at Ai2 before founding Trillium Labs.

The other side

Gary Marcus is AI's loudest skeptic, and following him keeps a feed honest. Yann LeCun (@ylecun), a Turing Award winner who left Meta to found AMI Labs, argues that LLMs alone will not reach human-level AI.

Turning posts into understanding

When a paper goes viral on X, look for the explainer before you form a view. ClaudeGPT Si works from primary sources, lists them under each story and labels explainers as such. Recent examples include DynaFlow trains a 27B model to design its own agent teams and the Ai2 study on unreliable novelty judges.

TechTwitter IO itself is a satirical daily paper about tech Twitter. It is good for a laugh about the week's AI drama, but it is parody, so get your actual research news from a source like ClaudeGPT Si.

When X alone is enough

If you are a working researcher already deep in one area, the right ten accounts and arXiv may be all you need. For everyone else, the posts move faster than the facts, and a sourced explainer saves time.