Скачать или смотреть GPT-5 Enters a Real Biology Lab and Boosts DNA Cloning Efficiency 79×

OpenAI has taken a major step beyond simulations: GPT-5 was tested inside a real biological wet lab — and it worked.

In collaboration with the startup Red Queen Bio, researchers ran the first experiment where a large language model actively participated in a closed-loop laboratory process. GPT-5 didn’t just analyze papers or suggest ideas in theory — it proposed hypotheses, planned experiments, received real lab results, and iterated based on physical outcomes.

The task was optimizing Gibson Assembly, a well-known and heavily studied DNA cloning protocol. Normally, even experienced biologists rarely optimize it deeply because results are case-specific and improvements usually top out at 2–3×.

GPT-5 surprised everyone. After several iterations, the system achieved a 79× increase in efficiency, measured by the number of successful colonies — and the result was stable and reproducible.
The key insight was adding two known proteins, RecA and gp32, into one reaction step. Both proteins are well-documented, but had never been functionally combined this way for cloning.

This isn’t a revolutionary breakthrough in biology — it’s closer to the level of a strong PhD student solving a narrow problem. But the real significance lies elsewhere:
GPT-5 successfully transitioned from being a text generator to an active participant in a physical scientific process.
That shift may be far more important than the 79× number itself.

#GPT5 #OpenAI #Biology #WetLab #DNACloning #GibsonAssembly #AIInScience #ArtificialIntelligence #Biotech #MachineLearning #LabAutomation #ScientificResearch #AIResearch #FutureOfScience #RedQueenBio