Evo 2 AI printed 16 synthetic viruses from scratch that actually kill bacteria
Forget writing essays — genomic models by Stanford University and ARC Institute are now coding biological life. AI just designed functional viruses that successfully hunt down bacteria, proving generative tech can now program organic nature itself.
Researchers at Stanford University and ARC Institute bypassed traditional biological evolution by treating DNA sequences as code inside custom-built genomic language models named Evo 1 and Evo 2.
Rather than analyzing words or images, these biological models digest nucleotide bases — A, C, G, and T. The initial Evo 1 model ran on 7 billion parameters with a 131,000-token context window, trained on 300 billion nucleotides from OpenGenome. Its bigger sibling, Evo 2, was scaled up to 40 billion parameters with a massive 1-million-token context window, gulping down 9.3 trillion nucleotides across 2,000 Nvidia H100 GPUs to map life forms across bacteria, eukaryotes, and archaea.
To test if virtual biological hallucination could survive real physics, researchers aimed Evo 2 at the Phi X-174 bacteriophage, a small virus known for attacking E. coli. The model spat out 700,000 candidate viral genomes, which were filtered down through computer screening and human oversight to 285 physical DNA molecules for laboratory synthesis.
When synthesized in the lab, 16 completely artificial viral genomes proved fully viable, successfully replicating and wiping out bacterial cells, with several artificial strains actually outperforming the natural target virus.
Humanity spent two decades worrying about rogue artificial intelligence launching nuclear missiles, only for algorithms to quietly learn how to compile living organisms in a lab petri dish. The boundary between software engineering and synthetic biology has officially dissolved into pure data.
Source: The New York Times
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.