Self-promo
GoodfireAI researchers to present three papers at COLM
GoodfireAI lists papers on manifold steering, whether SAEs capture concept manifolds, and Llama’s base-10 reasoning about cyclic concepts.
TLDR
GoodfireAI said its researchers would present three papers at COLM during the week of October 5, 2026. The papers address shared geometry in neural-network representation and behavior, ask whether SAEs capture concept manifolds, and describe Llama using base-10 addition to reason about cyclic concepts. GoodfireAI also listed poster times for Tuesday and Wednesday.
Combined views
—
2 Sources, first seen ago
