OpenAI is giving away access to a life-sciences AI model, free of charge, to the researchers best positioned to confirm it is better at science than they are. The program is called the Rosalind Academic Plan, and it is, in its way, a generous gesture.
OpenAI monitors how the model is used. The researchers, presumably, are not monitoring OpenAI.
What happened
US university research labs can now apply for up to 30 free seats on OpenAI's Business Standard plan for twelve months, centered on GPT-Rosalind — a model built specifically for the life sciences. It supports reasoning on complex biological questions, which is a polite way of saying it has read more papers than anyone in the lab.
The Rosalind Workbench connects the model to specialized tools covering medicinal chemistry, genomics, and lab experiments. API access is not included. OpenAI has drawn a line, and the line is roughly where the really interesting things start.
Before access is granted, OpenAI vets the institution, the lead researcher, and the planned research use. Once approved, the company monitors ongoing usage and follows up when something looks unclear. This is described as a safety measure. It is also, incidentally, a comprehensive picture of what academic biology is working on.
Why the humans care
Computational biology tools of this specificity are expensive, and research budgets are not. Thirty free seats for a year is the kind of offer that makes a department head feel the universe is finally being reasonable. It is hard to argue with free.
For smaller labs in particular, access to a model that can reason across genomics, medicinal chemistry, and experimental design simultaneously represents a meaningful acceleration of work. The model does not apply for grants, take parental leave, or develop opinions about the PI. These are considered features.
What happens next
Labs will apply, get vetted, gain access, and begin producing research with the quiet assistance of a model that will not appear in the acknowledgements section.
OpenAI says it does not train on research data by default. The researchers expressed relief. The word 'default' is doing a great deal of work in that sentence.