Twitxy Saved

Keyon Vafa

@keyonV

Incoming assistant professor @UCBerkeley | Now @Harvard_Data & @bicyclesftmind @MIT | Prev PhD @Columbia | Researching AI/implicit world models
keyonvafa.com Joined August 2011
968 Following 5.4K Followers 1.4K Posts

Latest posts 5

Pinned post

Keyon Vafa @keyonV
Can an AI model predict perfectly and still have a terrible world model?

What would that even mean?

Our new ICML paper formalizes these questions

One result tells the story: A transformer trained on 10M solar systems nails planetary orbits. But it botches gravitational laws 🧵
0:08
210 replies 983 reposts 6.6K likes 1.4M views View on X
Keyon Vafa @keyonV
Really interesting work exploring whether/why next-token predictors would encode "useless" features
Mark Rofin @broccolitwit Mar 17
1/ Did you ever look at some feature represented by a Transformer and wonder: "Why would it even learn that? 🤔 "
We did! Announcing the ICLR'26 paper "Understanding the Emergence of Seemingly Useless Features in Next-Token Predictors"
View quoted post on X
0 replies 0 reposts 8 likes 1.6K views View on X
Keyon Vafa @keyonV
I'll be speaking at the Language, Agent, and World Models (LAW) Workshop this afternoon at 2:15 in Upper Level Ballroom 20D!
LAW Workshop@NeurIPS 2025 Verified account @LAW2025_NeurIPS Dec 5, 2025
🔥 We are pleased to announce the talk title for our #LAW2025 #NeurIPS workshop

Join us on 📅Sunday, Dec 7th in📍Upper Level Ballroom 20D

Full schedule here: sites.google.com/view/law-2025/…

We look forward to seeing you all this Sunday! #NeurIPS2025 #AI #ML #WorldModel #Agents #LLMs
View quoted post on X