Claude secretly learned proteins like a biologist
Original headline: Claude was trained on protein sequences at scale — xinmingtu.cn
More in AI and machine learning news
Researchers discovered that Anthropic's latest Claude AI models, specifically Claude Fable 5.2 and Claude Opus 5.5, were trained on protein sequences at a massive scale. While other frontier AI models like those from OpenAI and Google show no sign of this training, Claude models can accurately predict the next amino acid in obscure bacterial proteins. The researchers tested this by giving models the beginning of poorly-studied proteins and asking them to complete the sequence, then comparing results to shuffled versions.
This capability is significant because Claude models outperform specialized open-source protein language models on these tasks, even matching ProtGPT3-10B, the largest dedicated protein model. The finding suggests Claude has internalized biological patterns from protein databases rather than just learning from scientific papers about proteins. This could transform how biologists use AI assistants for protein engineering and discovery, turning general-purpose chatbots into powerful bioinformatics tools.
Read the original at xinmingtu.cn
Satirical commentary on real tech news, summarised with AI. Read the original source for the facts. How ZeroDaily works
