OpenAI has created a language model that generates proteins to transform regular cells into stem cells, outperforming human efforts in this area. This development is the company's first focused on biological data and marks its initial public assertion that AI can yield unexpected scientific advancements, a major milestone in the pursuit of artificial general intelligence (AGI). OpenAI CEO Sam Altman has expressed confidence in his company's capability to develop AGI, asserting that advanced AI could greatly enhance scientific innovation. This protein engineering initiative began when Retro Biosciences, a San Francisco-based longevity research firm, sought collaboration with OpenAI. Notably, Altman has invested $180 million in Retro. Retro aims to extend human lifespan by ten years by researching Yamanaka factors—proteins that can revert skin cells to a stem cell-like state, capable of becoming any tissue type. However, current methods for reprogramming cells are inefficient, with a success rate below 1%. The new model, GPT-4b micro, was developed to enhance the efficacy of Yamanaka factors. Early results indicate that the model's modifications have made these factors over 50 times more effective. OpenAI researchers, including John Hallman and Aaron Jaech, are leading this initiative.

Although the findings await publication, the model's use remains limited to demonstration purposes for now. Unlike Google's AlphaFold, which predicts protein structures, GPT-4b was trained on diverse protein sequences and their interactions. Retro scientists utilized the model to propose redesigns of Yamanaka proteins, employing a prompting technique similar to "few-shot" learning. While traditional genetic engineering has limitations in the number of variations it can test, OpenAI's model often suggests numerous changes at once. Early results from Retro have confirmed that the model provides highly effective redesign proposals. An aging researcher from Harvard highlighted the importance of optimizing stem cell production, particularly for challenging cell types and species. However, the exact methodology behind GPT-4b's conclusions remains unclear, mirroring the complexities seen in other AI models. Though OpenAI states there were no financial exchanges in the collaboration, the partnership with Retro, which benefits from Altman's backing, raises questions about potential conflicts of interest. Altman's other investments in tech startups have sparked concerns about transparency and conflicts surrounding OpenAI's operations. Nonetheless, OpenAI maintains that its decision-making is independent of Altman’s personal investments.