Get artem kirsanov on Dwarkesh Podcast

Artem Kirsanov just published The Geometry of Prompting, a NAACL 2025 Findings paper with Kyunghyun Cho and SueYeon Chung that dissects how different prompting methods cause distinct, measurable shifts in language model representations. With frontier labs racing to deploy agentic systems, somebody who can explain the actual mechanics of in-context learning and task interference is must-hear today.

He also entered the Harvard PhD Program in Neuroscience for 2025–2026, bringing a rare bridge between neural circuit learning in brains and representation learning in large models at exactly the moment we need cross-pollination between AI and neuroscience. To our knowledge he has not been on the Dwarkesh Podcast before, and a scan of the show’s feed shows no prior episode with him, so this would be fresh territory for Dwarkesh’s audience.

Support this nomination today so listeners can get a timely, first-principles breakdown of how prompting really works and what it means for safer, smarter AI.