THE COLLECTION

EVERY SIGNAL, SORTED.

Search the whole shelf by topic, tool, term, or plain old curiosity.

SHOWING 1 OF 535 STORIES
ISSUE 589MORNING

United States, Allen Institute for AI, Ai2, StateBench, Long Context, State Tracking, Transformers, Recurrent Models, Hybrid Models, Gated DeltaNet, Model Evaluation, Software Systems, Data Provenance and Reliable AI

StateBench watched long documents change under the model. Test the update chain

A new benchmark separates finding a fact from carrying its changing state through thousands of updates. Frontier models handled the easy versions, then slipped badly when later facts depended on earlier ones.

#long-context#evaluation#model-architecture