kapynResearch

Recursive Language Models Meet Uncertainty: The Surprising Effectiveness of Self-Reflective Program Search for Long Context

New research explores self-reflective program search for long context in language models. This work investigates how to effectively select programmatic interactions that enable models to reliably extract and reason over information in extended contexts. The findings could significantly improve LLM performance on tasks requiring deep understanding of lengthy documents.

Apple ML Research·Jul 9, 2026

Opening Kapyn…