EMNLP 2025 Paper: Quantifying 'learning' in In-Context Learning via ICL ciphers

hanjie_chen · x · 2026-08-22

A paper accepted to the main conference of EMNLP 2025 investigates the extent to which Large Language Models (LLMs) can solve unseen tasks via In-context Learning (ICL).

The authors propose a new general framework called ICL ciphers to quantify "learning" in ICL via substitution ciphers.

Original post →

More from Research

Research channel →