9030 club reading "Evo2": #59 - ML Paper Reading Group

☕️📝 Paper Link 📝☕️ https://doi.org/10.1038/s41586-026-10176-5 Abstract (summary); Evo 2 introduces a biological foundation model trained on 9 trillion DNA base pairs from a highly curated genomic atlas spanning all domains of life. With a 1 million token context window at single-nucleotide resolution, the model accurately predicts the functional impacts of genetic variation—from noncoding pathogenic mutations to clinically significant variants—without task-specific fine-tuning. Evo 2's generative abilities can produce mitochondrial, prokaryotic, and eukaryotic sequences at a genome scale with unprecedented naturalness and coherence. By open-sourcing the model parameters, training code, and dataset, Evo 2 paves the way for accelerating the exploration and design of biological complexity. The Vibe; We find that having community leads to longer lives. We also love AI/science, so lets discuss papers together! Join us at Mox to explore: What are the biggest bottlenecks preventing genome-scale design from progressing as quickly as software, and does a massive biological foundation model…