What matters in AI.

Subscribe

Self distillation can decrease the correct answers of an AI model

The authors add Representation Geometry Regularization, which keeps the geometry of hidden representations and usually stops the decrease.

Claimed, not confirmed

Sources

  1. Mitigating Capability Degradation in LLM Self-Improvement via Representation Geometry Regularizationarxiv.org
AI MATTER · NEWS · AI MATTER · NEWS ·12 OCT2026

Posted

Tags

Learn the terms in this story

More in Research

All Research news