OpenAI announces GPT-Rosalind, a biology-focused AI model

In short: OpenAI says it built GPT-Rosalind, a version of its AI model trained to help with common biology research tasks, and it is starting with limited access.

What happened

OpenAI announced a new large language model called GPT-Rosalind, named after scientist Rosalind Franklin. A large language model is the kind of AI behind chatbots, trained to work with text and instructions (like an autocomplete that can follow directions).

OpenAI says GPT-Rosalind was trained on about 50 common biology workflows, plus how to use major public biology databases. The company says the goal is to help researchers deal with two problems, huge amounts of biological data and the fact that biology is split into many narrow specialties with their own terms and methods.

In a press briefing, OpenAI said the model can suggest likely biological pathways and help rank possible drug targets. A “drug target” is something in the body, often a protein, that a medicine might try to change.

OpenAI also said it tuned the model to be more skeptical, so it is more likely to say when an idea is not a good drug target. The report notes that it is still unclear how well OpenAI has addressed “hallucinations,” which is when an AI confidently makes up incorrect information.

For now, OpenAI is limiting who can use GPT-Rosalind. The company cited concerns about harmful use, such as helping someone make a virus more infectious. Only US-based organizations can apply for access through a trusted program, and OpenAI says it will tightly control availability. OpenAI also plans a more limited “Life Sciences Research Plugin” that will be broadly available.

Why it matters

AI tools are increasingly being aimed at specific jobs, not just general chat. If GPT-Rosalind works as described, it could help scientists search and connect information faster, but its accuracy and safety controls will matter just as much as its speed.

Source: Arstechnica