Short contexts maintain LLM performance on long-horizon tasks
Frontier language models sacrifice little to no task accuracy when maintaining short contexts across long-horizon tasks.
Stories per day · 30 days
1 stories, newest first · one card per story