AlphaEvolve has optimized one of it's own core transformer algorithms by 32%. Also, over the last year, Google has silently used AlphaEvolve it to optimize it's own TPU processors and their Borg database system.
Since it evolves to increase it's problem solving ability without human reinforcement, it can discover novel solutions on its own. To test that, Google gave it a slew of the hardest open math problems and it discovered new optimal solutions for 20% of them, matching humans on 75% and presumably giving less optimal solutions for 5%.
To investigate AlphaEvolve’s breadth, we applied the system to over 50 open problems in mathematical analysis, geometry, combinatorics and number theory. The system’s flexibility enabled us to set up most experiments in a matter of hours. In roughly 75% of cases, it rediscovered state-of-the-art solutions, to the best of our knowledge. And in 20% of cases, AlphaEvolve improved the previously best known solutions, making progress on the corresponding open problems.
Google chairman, Eric Schmidt, in a recent TEDtalk openly declared the AI race boils down to being first to develop a self-improving AI and then dominating all competition through the positive feedback loop of AI researching AI researching AI, etc... In hindsight he was hinting at AlphaEvolve, the timing was clearly not coincidental. For some reason he brought up the idea of bombing foreign data centers if they pulled ahead, but for now the US seems to still be far ahead and this so-called 'finish line' is now much closer than previously thought.
So what's the plan here? Personally I'm looking into biotech and pharma since drug discovery (including gene therapies) is going to accelerate and then automated organoid labs + AI research agents could blast through stage 3 trials (the real barrier), faster, safer, and more ethically.