ICL emerges across language, genomes, proteins, and more — the Convergent Emergence Hypothesis
DanielKhashabi · x · 2026-09-17
A new paper tests whether in-context learning is unique to LLMs by training models on six very different data types — language, genomes, proteins, integer sequences, time series, and images — on the same abstract tasks:
- Few-shot ICL emerges in all six modalities
- Surprisingly, across five modalities the same tasks tend to be easy or hard to learn in context, despite distinct per-modality strengths
- The authors propose the Convergent Emergence Hypothesis: ICL arises from next-token prediction across rich, structured data, with a shared core transcending modality
Work led by Nate Schlueter.
Related event: JHU study finds in-context learning emerges across six modalities(4 posts)→
More from Research
- Epoch AI: GPT-5.6 latency scales quadratically with context, Claude 5 stays near-linear — dl_weekly · 2026-09-17
- David Chalmers cites Jacques Thibodeau's work in his latest paper — JacquesThibs · 2026-09-17
- Insilico Medicine unveils 'Longevity Vaccines' combining AI and circular mRNA — rand_longevity · 2026-09-17
- NVIDIA's Axolotl3D does occlusion-aware 3D shape completion from multimodal inputs — NVIDIAAI · 2026-09-17
- alphaXiv maps every AI researcher's coauthorship graph to compute your LeCun number — burny_tech · 2026-09-17
- Gensyn releases open-1b, the first fully auditable and replayable foundation model — flngr · 2026-09-17