SignalTech/AI
Claude was trained on protein sequences at scale
Claude's training on protein sequences signals AI expansion into biotech applications.
Context
What happened: A next-amino-acid test on little-studied proteins finds that the latest Claude models were trained on protein sequences at scale, while other frontier models show no sign of it.
Key numbers: Fable 5.2 predicts the next amino acid correctly 54% of the time on real proteins versus 9% on shuffled ones.
Why it matters: The authors conclude protein sequences were a sizeable part of the models' training, and that GPT and Gemini did not see enough sequence data to learn these families.
What changed
A next-amino-acid test on little-studied proteins finds that the latest Claude models were trained on protein sequences at scale, while other frontier models show no sign of it.
Why it matters
The authors conclude protein sequences were a sizeable part of the models' training, and that GPT and Gemini did not see enough sequence data to learn these families.