Why Fine-Tuned Memories Still Struggle to Generalize
burny_tech · x · 2026-07-13
This paper examines a common phenomenon: LLMs can quickly memorize new facts during fine-tuning, but generalize poorly when multi-hop reasoning is required.
The author suggests the issue isn't "failure to learn knowledge," but a misalignment between knowledge and reasoning circuits. While facts exist inside the model, they don't reside in the layers actively involved in reasoning. By using self-patching to move appropriate hidden representations to the correct layers, the paper restored a significant portion of generalization ability; a simple heuristic recovered 58–75% of the oracle headroom.
Related event: Study Reveals Knowing-Using Gap in LLM Fine-Tuning(2 posts)→
More from Models
- Cohere Flags Qwen Sandbox Oddity; Security Researchers Call It FUD — basedjensen · 2026-10-27
- GPT-6.1 Terra Now Testable on LMArena in Battle and Agent Modes — arena · 2026-10-27
- OpenAI Launches GPT-6.1 Terra at $1/M Input Tokens, ~9x Cheaper per Task Than Opus 5.5 — VraserX · 2026-10-27
- GPT-6.1 Terra Benchmarks: Sweeps All Six Tests, SWE-bench at 76.2% — mark_k · 2026-10-27
- GPT-6.1 Terra lands: early users praise speed, value and subagent performance — BenBajarin · 2026-10-27
- Leaked GPT-6.1 Terra benchmarks: cheap but 'boring', still far behind Astra and Claude — Angaisb_ · 2026-10-27