Google’s AI research division, DeepMind, has unveiled a groundbreaking AI system called AlphaEvolve, designed to tackle complex mathematical and scientific problems, as reported by TechCrunch on Wednesday.
AlphaEvolve is set to revolutionize Google’s AI training infrastructure. The system introduces an innovative automatic evaluation process to mitigate hallucination issues prevalent in existing AI models. According to company representatives, this self-review mechanism allows the AI to scrutinize its own responses and assess their accuracy, significantly boosting reliability.
DeepMind asserts that AlphaEvolve, leveraging the latest Gemini model, outperforms previous AI solutions by a considerable margin.
To function, AlphaEvolve requires users to input problems and provide formulas for automatic evaluation. This specificity limits its application to fields such as mathematics, computer science, and systems optimization, making it less suitable for non-numerical problem-solving.
DeepMind reports that AlphaEvolve has already been put to practical use, optimizing computing resources in Google’s data centers and reducing AI training times.