Study finds Claude was trained on protein sequences at scale

gmays · hn · 2026-11-18

A blog post analyzes evidence that Claude (likely Opus) was trained on large-scale protein amino acid sequences. The author shows the model can parse FASTA-format protein data, reason about sequence structure, and even generate plausible sequences — suggesting scientific sequence data may be a hidden component of frontier LLM pretraining corpora, with implications for biology–AI crossover research.

Original post →

More from Models

Models channel →