Toy transformers may represent belief-state geometry optimally but not minimally

·LessWrong··

Methods note: The code used for the experiments and related open-source repo were built with Claude. The experimental design and writeup is my own, with minimal editing and formatting amendments made with Claude. Thesis A toy transformer keeps provably predictively defunct belief state data in its residual stream. This information is shed only when there is a sufficient amount of imposed capacity pressure, in which case the oldest predictively defunct information is shed first. Results I conduct...

Read full article →

Related Articles

AI's Affordability Crisis
ilreb · Hacker News · 15h ago
Trains halted across Germany because of communication system problem
sva_ · Hacker News · 8h ago
Algorithmic Monocultures in Hiring
sizzle · Hacker News · 11h ago
F3
tosh · Hacker News · 13h ago
75% More Pedestrians Have Been Killed Since 2009. Giant Trucks and SUVs Are Why
theanonymousone · Hacker News · 12h ago