News Summary

DeepMind has developed a new AI system named AlphaEvolve aimed at solving machine-gradable problems, particularly in math and science domains. The system assists in optimizing Google’s infrastructure, notably improving data center efficiency and speeding up model training times. AlphaEvolve introduces an automatic evaluation mechanism to reduce hallucinations, which are common in probabilistic AI models. It has demonstrated success in rediscovering known solutions and uncovering new ones in a significant percentage of tested math problems. While not making groundbreaking discoveries, AlphaEvolve enhances efficiency, allowing experts to concentrate on more critical tasks.