At ICML, Petar Veličković presents study on how large language models compute verbal confidence
The research used comparative evaluations on PaLM and Gemini.
Story Brief
The research used comparative evaluations on PaLM and Gemini.
Commentary on X
Highest ranked
Azmine Wasi ✈️ ICML 2026@AzmineWasiTECH
@PetarV_93 @dharshsky @ArthurConmy @fedzbar @sindero Would love to meet you, and maybe take a picture too! I’m following your work since 2022, back when I worked on GNNs. The last talk of yours I attended was on improving LLMs with GNNs. You asked which token will the LLM remember? That question changed how I write prompts.:)
Petar Veličković@PetarV_93
