Google shows off AlphaEvolve, an AI system that fixes up codes made by a similar system called Gemini, making programming more suitable and less likely to make mistakes.

Ahead of the highly expected Google I/O 2025 conference, Google has come out with AlphaEvolve, a new AI that aims to make finding and improving computer algorithms much faster and better. Developed by Google DeepMind, AlphaEvolve uses Gemini 2.0 large language models to help solve tough coding problems by slowly improving them in a step-by-step process.

Unlike regular code generators, AlphaEvolve looks at each new code solution and keeps the better ones while getting rid of the ones that aren’t as good, which helps it come up with better code over time. According to Google, they’ve already used this new method to improve their job scheduling system, which helped cut down on energy use by 0.7% across all their data centers around the world.

Beyond just being better at tasks, AlphaEvolve also works to check for things like fake information by using an automated system to keep things honest. This process helps make sure that the best and most correct answers stick around, so the tool works really well when you need to be precise, like in computer science or math.

Google DeepMind says that AlphaEvolve is helpful in situations where you can check the answers by using specific set rules or ways of testing. However, the system can only do certain kinds of tasks that are already laid out in programming codes and isn't helpful for questions that need more creativity or judgment.

As Google keeps pushing the limits with AI, the AlphaEvolve software is a big step toward making AI systems that people can really trust and that keep getting better on their own.