AlphaEvolve - takes me back into - my - history and into a bright future ...
AlphaEvolve - takes me back into - my - history and into a bright future …
This month Google DeepMind released AlphaEvolve a new AI Agent which evolves algorithms for math and practical applications. In its core it’s combining evolutionary algorithms(SA) with the power of LLM.
Back in university, ouch we talk like 25y back, I studied evolutionary algorithms. EAs are in short an optimization techniques inspired by natural evolution . I tried to prove upper and lower bounds and the positive impact of providing and maintaining diversity in the populations. Hard work, with very limited success on my end. Since then I really looked back into EAs, but they are more relevant than ever.
In AlphaEvolve EAs are combined with current SOTA LLMs to evolve code for arbitrary problems with great success. Solving real world problems and finding new, improved algorithms in fields where human have been researching for decades already. Really impressive.
And this is “just” one of many Agents in the upcoming world of Agentic AI. Imagine the potential of combing this with Agents of other capabilities.
I suggest to have a read of the blog post [1] and possible dive deep into the in the whitepaper [2]. I took a guided tour via the MLST podcast [3] as well with Matej Balog and Alexander Novikov. Super insightful and highly recommended.
Disagree and Commit: In order to satisfy the annoying and short-sided algorithm on this very platform hosting this post, you find the very helpful, but external to the platform, links in the comments. Avoiding any undifferentiated heavy lifting on your side.
Cross-posted to LinkedIn