Announcement
LLMs appear to learn novel tasks more reliably from rules than examples
The authors say their EMNLP 2026 paper compares in-context learning from rules and examples across diverse tasks and models.
TLDR
The authors of an EMNLP 2026 paper say they compared how LLMs learn novel tasks in context from rules versus examples across diverse tasks and models. They report that LLMs generally learned more reliably from rules.
Combined views
3.1K
3 Sources, first seen ago
