Three COLM posters planned on number representations, off-policy learning and reasoning
A researcher said they planned to present three posters at COLM in the Bay Area the week of October 5, covering how different models learn similar number representations, token-level off-policy learning and SAEs for reasoning.
TLDR
On September 28, a researcher said they were on the job market and planned to present three posters at COLM in the Bay Area the following week. The topics are how different models learn similar number representations, token-level off-policy learning and SAEs for reasoning.
Combined views
1 Source, first seen 4h ago
3 reposts