DeepMind

DeepMind's AlphaFold cracked the 20-year protein folding problem

Shaan explains that for ~20 years a global protein-folding contest topped out around 60/100 in accuracy; this year DeepMind's AI hit ~90 by training on existing folds, a breakthrough that could let us design proteins to fight disease and edit the body.

“And then this year, DeepMind got to like 90, and it was like DeepMind basically broke through. They won the special award for winning the competition. And the way that they broke through was because they basically trained their machine learning algorithms to— on existing folding sequences to learn what a new sequence might fold— look like folded.”