Study Examines Where Distilled Student Models Store Long-Form Reasoning
New arXiv research finds small student models degrade when learning from large reasoning models' long solutions, prompting methods that shorten them by interpolating an LRM with a short-answer model.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-04 05:00 · arXiv cs.CL
Where Do Distilled Students Store Thinking?