Google’s AlphaEvolve Is Evolving New Algorithms — And It Could Be a Game Changer

Models have undeniably revolutionized how many of us approach coding, but they’re often more like a super-powered intern than a seasoned architect. Errors, bugs and hallucinations happen all the time, and it might even happen that the code runs well but… it’s not doing exactly what we wanted.

Now, imagine an AI that doesn’t just write code based on what it’s seen, but actively evolves it. To a first surprise, this means you increase the chances of getting the right code written; however, it goes far beyond: Google showed that it can also use such AI methodology to discover new algorithms that are faster, more efficient, and sometimes, entirely new.

I’m talking about AlphaEvolve, the recent bombshell from Google DeepMind. Let me say it again: it isn’t just another code generator, but rather a system that generates and evolves code, allowing it to discover new algorithms. Powered by Google’s formidable Gemini models (that I intend to cover soon, because I’m amazed at their power!), AlphaEvolve could revolutionize how we approach coding, mathematics, algorithm design, and why not data analysis itself.

How Does AlphaEvolve ‘Evolve’ Code?

Think of it like natural selection, but for software. That is, think about Genetic Algorithms, which have existed in data science, numerical methods and computational mathematics for decades. Briefly, instead of starting from scratch every time, AlphaEvolve takes an initial piece of code – possibly a “skeleton” provided by a human, with specific areas marked for improvement – and then runs on it an iterative process of refinement.

Let me summarize here the procedure detailed in Deepmind’s white paper:

Intelligent prompting: AlphaEvolve is “smart” enough to craft its own prompts for the underlying Gemini Llm. These prompts instruct Gemini to act like a world-class expert in a specific domain, armed with context from previous attempts, including the points that seemed to have worked correctly and those that are clear failures. This is where those massive context windows of models like Gemini (even you can run up to a million tokens at Google’s AI studio) come into play.

Creative mutation: The LLM then generates a diverse pool of “candidate” solutions – variations and mutations of the original code, exploring different approaches to solve the given problem. This parallels very closely the inner working of regular genetic algorithms.

Survival of the fittest: Again like in genetic algorithms, but candidate solutions are automatically compiled, run, and rigorously evaluated against predefined metrics.

Breeding of the top programs: The best-performing solutions are selected and become the “parents” for a next generation, just like in genetic algorithms. The successful traits of the parent programs are fed back into the prompting mechanism.

Repeat (to evolve): This cycle – generate, test, select, learn – repeats, and with each iteration, AlphaEvolve explores the vast search space of possible programs thus gradually homing in on solutions that are better and better, while purging those that fail. The longer you let it run (what the researchers call “test-time compute”), the more sophisticated and optimized the solutions can become.

Building on Previous Attempts

AlphaEvolve is the successor to earlier Google projects like AlphaCode (which tackled competitive Programming) and, more directly, of FunSearch. FunSearch was a fascinating proof of concept that showed how LLMs could discover new mathematical insights by evolving small Python functions.

AlphaEvolve took that concept and “injected it with steroids”. I mean this for various reasons…

First, because thanks to Gemini’s huge token window, AlphaEvolve can grapple with entire codebases, hundreds of lines long, not just tiny functions as in the early tests like FunSearch. Second, because like other LLMs, Gemini has seen thousands and thousands of code in tens of programming languages; hence it has covered a wider variety of tasks (as typically different languages are used more in some domains than others) and it became a kind of polyglot programmer.

Note that with smarter LLMs as engines, AlphaEvolve can itself evolve to become faster and more efficient in its search for solutions and optimal programs.

AlphaEvolve’s Mind-Blowing Results on Real-World Problems

Here are the most interesting applications presented in the white paper:

• Optimizing efficiency at Google’s data centers: AlphaEvolve discovered a new scheduling heuristic that squeezed out a 0.7% saving in Google’s computing resources. This may look small, but Google’s scale this means a substantial ecological and monetary cut!
• Designing better AI chips: AlphaEvolve could simplify some of the complex circuits within Google’s TPUs, specifically for the matrix multiplication operations that are the lifeblood of modern AI. This improves calculation speeds and again contributes to lower ecological and economical costs.
• Faster AI training: AlphaEvolve even turned its optimization gaze inward, by accelerating a matrix multiplication library used in training the very Gemini models that power it! This means a slight but sizable reduction in AI training times and again lower ecological and economical costs!

• Numerical methods: In a kind of validation test, AlphaEvolve was set loose on over 50 notoriously tricky open problems in mathematics. In around 75% of them, it independently rediscovered the best-known human solutions!

Towards Self-Improving AI?

One of the most profound implications of tools like AlphaEvolve is the “virtuous cycle” by which AI could improve AI models themselves. Moreover, more efficient models and hardware make AlphaEvolve itself more powerful, enabling it to discover even deeper optimizations. That’s a feedback loop that could dramatically accelerate AI progress, and lead who knows where. This is somehow using AI to make AI better, faster, and smarter – a genuine step on the path towards more powerful and perhaps general artificial intelligence.

Leaving aside this reflection, which quickly gets close to the realm of science function, the point is that for a vast class of problems in science, engineering, and computation, AlphaEvolve could represent a paradigm shift. As a computational chemist and biologist, I myself use tools based in LLMs and reasoning AI systems to assist my work, write and debug programs, test them, analyze data more rapidly, and more. With what Deepmind has presented now, it becomes even clearer that we approach a future where AI doesn’t just execute human instructions but becomes a creative partner in discovery and innovation.

Already for some months we have been moving from AI that completes our code to AI that creates it almost entirely, and tools like AlphaFold will push us to times where AI just sits to crack problems with (or for!) us, writing and evolving code to get to optimal and possibly entirely unexpected solutions. No doubt that the next few years are going to be wild.

References and Related Reads

www.lucianoabriata.com I write about everything that lies in my broad sphere of interests: nature, science, technology, programming, etc. Subscribe to get my new stories by email. To consult about small jobs check my services page here. You can contact me here. You can tip me here.