OpenAI decided not to create a universal model but to focus on a specialized tool for biology. The company introduced GPT-Rosalind, a powerful language model trained on key biological processes and extensive databases. Its name is a tribute to Rosalind Franklin, a prominent figure in molecular biology. ## Deep Dive into Biology Developers claim that GPT-Rosalind is designed to address the scale and complexity challenges faced by modern biology. Over decades of genome and protein biochemistry research, scientists have amassed colossal volumes of data that are impossible for a single person to grasp. Furthermore, biology has split into numerous narrow fields, each with its unique language and methods. For example, a geneticist studying gene function in brain cells quickly encounters a vast body of neurobiological research that requires additional assistance. OpenAI trained the model on the 50 most common biological processes and also taught it to reference key public scientific sources. ## Capabilities of GPT-Rosalind Thanks to such deep tuning, GPT-Rosalind can suggest probable biological pathways and link genotype to phenotype through known regulatory mechanisms. It can also draw conclusions about potential structural or functional properties of proteins and effectively assist in selecting potential targets for new drugs. The company emphasizes that it has worked to mitigate the typical problem of language models—an excessive willingness to agree with users and provide overly confident advice. GPT-Rosalind was made more skeptical to more frequently point out weak hypotheses and dubious targets for drug development. OpenAI also claims that the system is capable of navigating complex multi-step reasoning chains, citing test results where the model demonstrated expert-level performance. However, the main question remains open: how reliably does GPT-Rosalind distinguish useful scientific conclusions from convincingly sounding errors. The issue of "hallucinations" remains one of the most pressing for large language models, and in scientific work, the cost of such an error is particularly high. Therefore, around the new system, almost certainly, there will quickly emerge both enthusiastic reviews of unexpected discoveries and examples of gross missteps, where the model suggests biologically meaningless actions. ## Dangers and Access Limitations For this reason, OpenAI is not opening access to everyone without restrictions. The company fears that such a specialized model could be misused, for example, for tasks related to increasing the contagiousness of viruses. Currently, only organizations from the U.S. can apply to work with GPT-Rosalind under a strictly controlled access scheme. OpenAI plans to further limit the circle of users of this powerful system. For a broader audience, the company promises to release a scaled-down plugin called the Life Sciences Research Plugin. ## Unique Approach of OpenAI There are already language models and agent systems for science on the market, but most of them aim to cover several disciplines at once. GPT-Rosalind looks different: OpenAI is betting specifically on biology and on deep tuning for the specific tasks of laboratories and research groups. How much this narrow specialization will provide a real advantage will only become clear after the first independent reviews and practical work with the model.