Predict protein structures better than any other methods.

Here we provide the first computational method that can regularly predict protein structures with atomic accuracy even in cases in which no similar structure is known. We validated an entirely redesigned version of our neural network-based model, AlphaFold, in the challenging 14th Critical Assessment of protein Structure Prediction (CASP14)15, demonstrating accuracy competitive with experimental structures in a majority of cases and greatly outperforming other methods.

They are very good pattern matching machines. Most of our life science scientists are out there finding patterns. Like which antibodies pair with which cellular components to do things like predict cancer before it is a problem.

They are also very good at determining locations for clinical trials based on criteria found in previous trials that would be near impossible to do for a human.

Valve is trying to use neural networks for their anti cheat. They want to move it entirely to server side and rely less on the client to make Linux gaming an industry standard.

Instead of spying on your PC to see if you’re using something you’re not supposed to, they want to examine your behavior instead and act based upon that. I think this is a good usecase.

Fixing my PC's rebooting issue, for one. It diagnosed the error logs that I gave it, and suggested to use my motherboard's Load-Line Calibration feature to prevent the shutdowns. Before, I could get over a dozen reboots in a day.

For me, not having sudden reboots gives me a great deal of mental peace.