When artificial intelligence and science intersect, the results can be groundbreaking. Consider AlphaFold, the protein-folding program by Google’s DeepMind, which earned a Nobel Prize for its creator and fundamentally transformed biology. Now, OpenAI has entered the scientific arena with its own unique contribution: a language model designed specifically for engineering proteins, named GPT-4b micro.
Unlike the more general-purpose ChatGPT models available to the public, GPT-4b micro has a singular focus—reengineering proteins to advance human longevity. Developed in collaboration with Retro Biosciences, a San Francisco-based biotech startup funded heavily by OpenAI CEO Sam Altman, the model represents OpenAI’s first foray into biological research. The partnership between the two organizations aims to tackle a grand challenge: adding a decade to the human lifespan.
Retro Biosciences’ work centers on Yamanaka factors, a set of proteins capable of transforming mature skin cells into young stem cells. Stem cells, in turn, can generate any type of tissue in the human body, offering hope for therapies ranging from organ regeneration to reversing cellular aging. However, reprogramming cells using Yamanaka factors is notoriously inefficient, taking weeks to complete and succeeding in fewer than 1% of treated cells.
The Role of GPT-4b Micro
This is where GPT-4b micro comes into play. Unlike Google’s AlphaFold, which predicts the 3D shapes of proteins, GPT-4b micro focuses on redesigning proteins to enhance their function. The AI model was trained on a curated dataset of protein sequences from various species, as well as data about protein-protein interactions. While this represents a smaller dataset compared to OpenAI’s flagship language models, GPT-4b micro is an example of a “small language model” designed for precision in a specialized domain.
Retro scientists used the model to generate redesigns of the Yamanaka proteins, leveraging techniques similar to “few-shot” prompting in traditional language models. This method allows users to guide the AI by presenting examples of desired outcomes before requesting suggestions. The AI’s outputs frequently included significant alterations to the proteins, with up to a third of their amino acids modified—changes that human researchers wouldn’t typically consider due to the vast number of possible combinations.
When tested in the lab, GPT-4b micro’s suggestions yielded remarkable results. The model proposed modifications that improved the effectiveness of two key Yamanaka factors by over 50 times, based on preliminary measures. According to Retro’s CEO Joe Betts-Lacroix, the AI’s suggestions were unusually effective, leading to substantial advancements that human scientists alone had not achieved.
A New Frontier for AI in Science
The success of GPT-4b micro highlights the growing role of AI in scientific discovery. John Hallman, an OpenAI researcher who helped develop the model, emphasized its ability to deliver results beyond human capability. Meanwhile, Vadim Gladyshev, a Harvard aging researcher consulting for Retro, noted that improved methods for creating stem cells would have far-reaching implications. “Skin cells are relatively easy to reprogram,” Gladyshev explained, “but for other cells or species, the challenges are immense. This technology could address those hurdles.”
The broader implications of this work could be transformative. While GPT-4b micro is not yet available as a commercial product, it demonstrates the potential of tailored AI models to accelerate research in highly specialized fields. It also raises questions about whether AI can achieve true scientific discovery—an important milestone on the path to artificial general intelligence (AGI).
Ethical and Controversial Ties
Despite the promise of GPT-4b micro, its development is not without controversy. Retro Biosciences approached OpenAI about the collaboration roughly a year ago, a connection that was anything but coincidental. Sam Altman, OpenAI’s CEO, had personally invested $180 million in Retro, as reported by MIT Technology Review in 2023.
Although OpenAI insists that no money exchanged hands for the collaboration and that Altman was not directly involved in the project, his dual role as Retro’s largest investor and OpenAI’s leader has drawn scrutiny. Critics point to the possibility of conflicts of interest, especially given Altman’s extensive private investments in tech startups, which some claim form an “opaque investment empire.”
OpenAI maintains that its decision-making process is independent of Altman’s personal ventures, but the association with Retro could boost the startup’s visibility and fundraising potential. Betts-Lacroix declined to comment on whether Retro is currently seeking additional investment.
The Path Ahead
Retro Biosciences’ work, bolstered by GPT-4b micro, is still in its early stages. While the company plans to publish its findings, external validation will be necessary to confirm the AI’s impact. Even so, the results are a promising glimpse into the future of biotech and AI collaboration.
The partnership also underscores the potential of bespoke AI models in addressing some of the world’s most complex challenges. As Altman recently noted, “Superintelligent tools could massively accelerate scientific discovery and innovation well beyond what we are capable of doing on our own.” If GPT-4b micro’s early success is any indication, OpenAI may be paving the way for a new era in both science and AI development.
While questions about ethics, access, and affordability remain, the prospect of AI-driven breakthroughs in longevity research is undeniably exciting. OpenAI’s latest experiment could represent the first steps toward not only extending human life but fundamentally redefining what it means to age.
