.png)
January 21, 2025
OpenAI, known for its advancements in the field of artificial intelligence, is now venturing into scientific research. With a new language model called GPT-4b micro, the company aims to revolutionize the production of stem cells. Unlike Google's Alphafold, which predicts protein folding, OpenAI's model takes a different approach. It focuses on the development of new proteins that can convert ordinary cells into stem cells.
This project not only marks OpenAI's first foray into biological data processing but also the company's first public claim to generate new scientific findings with its models. This raises the question of whether AI can independently make scientific discoveries in the future – a potential milestone on the path to Artificial General Intelligence (AGI).
The development of the protein engineering project began with a collaboration between OpenAI and Retro Biosciences, a company dedicated to longevity research. OpenAI CEO Sam Altman has personally provided Retro Biosciences with substantial funding. The shared goal is to extend the human lifespan by at least ten years. Retro Biosciences is researching the so-called Yamanaka factors, a group of proteins that can convert skin cells into stem cells. These stem cells have the potential to produce any tissue in the body.
Reprogramming cells with the Yamanaka factors is currently an inefficient process. OpenAI's new model, GPT-4b micro, has been trained to provide suggestions for optimizing these protein factors. Initial results indicate that the modified factors, based on the model's suggestions, are significantly more effective than the original Yamanaka factors.
The suggestions from GPT-4b micro have already been tested in laboratory experiments and show promising results. Although the publication of detailed results is still pending, it is planned to make them available to the scientific community. The model itself is not yet publicly available. It is a custom-made demonstrator whose future integration into OpenAI's product range remains open.
Unlike Google's Alphafold, which predicts the folding of proteins, GPT-4b micro takes a different approach. Since the Yamanaka factors are unstructured proteins, the language model is better suited for their analysis and optimization. GPT-4b micro has been trained with a large amount of data, including protein sequences from various species and information about protein interactions. Although the dataset is smaller compared to the training data of OpenAI's chatbots, it still delivers valuable results.
The scientists at Retro Biosciences use a prompting method, similar to the "few-shot" method in chatbots, to control GPT-4b micro and obtain suggestions for protein modifications. The model often generates suggestions with extensive changes to the amino acid sequence, which go beyond the capabilities of conventional genetic engineering methods.
As with many AI models, the exact workings of GPT-4b micro remain unclear in detail. Researchers are working to better understand the mechanisms behind the model's suggestions. This is crucial to fully exploit the model's potential and further optimize its application in research.
The connection between Altman, OpenAI, and Retro Biosciences raises questions about potential conflicts of interest. Altman's investments in various tech startups, including Retro Biosciences, could draw criticism, especially since some of these companies collaborate with OpenAI. However, OpenAI emphasizes that Altman is not directly involved in the research work and that business decisions are made independently of his private investments.
Sources: - https://t3n.de/news/laenger-leben-dank-ki-openai-baut-ein-sprachmodell-fuer-biologische-daten-1669026/ - https://t3n.de/news/ - https://t3n.de/news/openai-mitgruender-sagt-fundamentalen-wandel-fuer-die-ki-entwicklung-voraus-1663792/ - https://neuron.expert/news/openai-has-created-an-ai-model-for-longevity-science/10405/de/ - https://t3n.de/news/nicht-nur-ki-agenten-und-kleine-sprachmodelle-diese-ki-trends-fuer-2025-solltest-du-kennen-1667074/ - https://www.youtube.com/watch?v=Gj0Rpes81J8 - https://t3n.de/news/warum-das-ki-modell-o3-von-openai-so-hohe-kosten-verursacht-und-welche-leistungen-es-verspricht-1666591/ - https://www.faz.net/pro/digitalwirtschaft/kuenstliche-intelligenz/liveticker-zu-12-days-of-open-ai-sam-altman-stellt-open-ais-neues-sprachmodell-o3-vor-faz-110155698.html - https://shop.t3n.de/products/t3n-76 - https://www.faz.net/podcasts/podcasts/neues-von-openai-wie-weit-ki-jetzt-wirklich-ist-110013229.html