OpenAI, in collaboration with Retro Biosciences, has created a model that can help rejuvenate animals, create human organs, or ensure the supply of replacement cells by transforming normal cells and stem cells.
OpenAI, in collaboration with Retro Biosciences, has created a model that can help rejuvenate animals, create human organs, or ensure the supply of replacement cells by transforming normal cells and stem cells.
The GPT-4b micro model the researchers created was trained to suggest ways to rework protein factors to improve their function. According to some preliminary results cited by the company, the researchers used the model’s suggestions to reengineer the proteins, two of Yamanaka’s factors. The results should make them more than 50 times more efficient, MIT Technology Review reports .
«Overall, these proteins seem to be better than what scientists could have made on their own,» says John Hollman, a researcher at OpenAI. In addition to Hollman, the model’s lead developers were Aaron Jech of OpenAI and Rico Meinl of Retro.
The model is not yet available for widespread use.
In October 2024, Google’s Nobel Prize- winning AlphaFold model predicted the shape of proteins. Since 2020, the tool has helped predict the structure of nearly all 200 million proteins identified by researchers.
Unlike Google’s model, GPT-4b micro was trained on protein sequence examples from many species, as well as information about which proteins tend to interact with each other. While that’s a lot of data, it’s only a fraction of what OpenAI’s flagship chatbots were trained on, making GPT-4b an example of a «small language model» that works with a focused dataset.
In 2023, Sam Altman, CEO of OpenAI, personally funded Retro with $180 million.
