ICL emerges across language, genome, protein, image and timeseries training, JHU study shows
DanielKhashabi · x · 2026-09-16
New JHU CLSP research, 'Convergent Emergence of In-Context Learning Across Modalities', shows models trained via next-token prediction on language, genomes, proteins, images, timeseries, and integer sequences all exhibit few-shot in-context learning — and their ICL performance is highly correlated across modalities.
Related event: JHU Study: In-Context Learning Emerges Across Modalities(2 posts)→
More from Research
- Microsoft and SJTU open-source Argus, an agent that ran 1,548 hours and solved a 20-year-old math problem — jiqizhixin · 2026-09-16
- Composing Continual Learning Mechanisms Boosts Long-Horizon Memorization in LMs — JohnsHopkins · 2026-09-16
- Gavel Elicits Native Skill Routing from a Frozen LLM's Hidden States — Tsinghua · 2026-09-16
- ScienceBuddy: Recursive-in-Recursive Self-Improvement for Scientific Agents — Shuhan Xue · 2026-09-16
- RCT: LLMs fail to significantly boost novices' wet-lab molecular biology success — shae_mcl · 2026-09-16
- Six years on, scvi-tools still widely used — outlasting foundation models and coding agents — anshulkundaje · 2026-09-16