BLISS AI Speaker Series: Understanding Reasoning from Pretraining to Post-Training — Pavel Izmailov

A 45-minute research talk from the BLISS AI Speaker Series, with Pavel Izmailov, Member of Technical Staff at Anthropic, on how reasoning emerges in language models: what pretraining puts there, and what post-training actually changes. Hosted by BLISS Berlin e.V. at TU Berlin's chemistry institute on Straße des 17. Juni. Talk first, then open discussion with Berlin's machine-learning community over free drinks and pizza. Aimed at students, PhD candidates, academic researchers and industry people doing ML research. RSVP on Luma is required for entry and doors close at 19:15. Photography may take place. Why go: reasoning is the capability every lab claims to have improved and almost nobody explains mechanistically. Izmailov works on it inside a frontier lab and is speaking to a research audience, not a buyer. Best for ML researchers and engineers working on training or evaluation.