EMNLP Paper Finds LLMs Learn Novel Tasks More Reliably From Rules Than From Examples
TuhinChakr · x · 2026-10-06
An EMNLP 2026 Main paper systematically compares in-context learning from rules versus examples across diverse tasks and models.
- Key finding: LLMs generally learn more reliably from rules than from examples
- The authors also observed predicted differences in learning modes for inferring conjunctive vs. disjunctive concept definitions
- Co-author Najoung Kim notes it's the kind of obviously-useful study that surprisingly didn't exist yet
Related event: EMNLP Paper: LLMs Learn New Tasks More Reliably from Rules than Examples(2 posts)→
More from Research
- New paper probes multimodal MoE expert structure to make model adaptation much faster — georgiagkioxari · 2026-10-06
- Developer Runs 1,200+ Blender Modeling Sessions to Blind-Rank LLMs and Agent Harnesses — Izolight · 2026-10-06
- FOVEATED Fixes Context Reliance in Unstructured Knowledge Editing Across 5 LLM Editors — Ding Wu · 2026-10-06
- Software-in-the-loop: self-supervised scaling of terminal environments lifts Terminal-Bench 2 to 53.56% — Zhongzhi Li · 2026-10-06
- CC-Bench at COLM 2026 finds LLMs still default to stereotypes over implicit cultural cues — MaartenSap · 2026-10-06
- UCLA PhD's LLM agents produce 126K-line Lean 4 proof of MIP* = RE core theorem in 63 days — siyan_zhao · 2026-10-06