AlphaEvolve: DeepMind’s AI That Builds Its Own Algorithms
AlphaEvolve: DeepMind’s AI That Builds Its Own Algorithms
DeepMind, the team behind famous AI projects like AlphaGo and AlphaFold, just introduced something new: AlphaEvolve. It’s an AI system that can come up with brand-new algorithms on its own. That’s like teaching a computer not just to follow steps, but to invent better steps.
What Is AlphaEvolve?
AlphaEvolve is powered by two things:
- A smart language model (like Gemini, which can understand and write code), and
- An evolution-like system that keeps testing and improving ideas until the best ones are found.
So, if you give AlphaEvolve a challenge—like “make this math faster”—it will keep trying different solutions until it finds a better way to do it.
What Can It Do?
Solve tough math problems
It improved a well-known problem in geometry called the “11-dimensional kissing number.” That’s something even expert mathematicians struggle with.
Find better code
It found a faster way to multiply certain types of numbers—beating a famous algorithm that’s been used since 1969.
Make real-world systems more efficient
AlphaEvolve was used in Google data centers and helped save about 0.7% of computing power. That’s a big deal at Google scale.
How It Works
- The AI generates lots of possible solutions
- It tests them all
- It keeps the good ones and improves them
- It repeats the process over and over
Why Should You Care?
This isn’t just tech for scientists. AlphaEvolve could one day:
- Help software run faster
- Assist with scientific research
- Reduce costs by improving how computers work
AlphaEvolve is a big step in AI development. It’s still growing, but it already shows that AI can help solve problems we didn’t know how to tackle before. This kind of innovation is worth watching. Read more here.