Image: Reuters
Alphabet's Google introduced two groundbreaking AI systems on Thursday, showcasing significant advancements in solving complex mathematical problems. This marks a notable achievement in the realm of generative AI, an area where previous AI models, which predict the next word statistically, have struggled, especially with abstract math requiring human-like reasoning.
DeepMind, Google’s AI division, revealed that their new AI models, AlphaProof and AlphaGeometry 2, successfully answered four out of six questions at the 2024 International Math Olympiad, a prestigious competition for high school students. This achievement underscores the models' progress in tackling advanced mathematical challenges.
In a blog post, Google mentioned that while one problem was solved in minutes, others took up to three days—longer than the competition’s time limit. Despite this, the results are the best any AI system has achieved in such a competition to date.
AlphaProof was developed to enhance reasoning capabilities by merging Gemini, the language model behind Google's chatbot, with AlphaZero, an AI system that has previously outperformed humans in games like chess and Go. AlphaProof managed to solve three of the competition’s problems, including the most difficult question, which only five out of over 600 human contestants could solve.
AlphaGeometry 2 tackled and solved an additional math problem, further demonstrating Google's progress in AI-driven mathematical reasoning.
Interestingly, Microsoft-backed OpenAI is also making strides in this field. It was reported earlier in July that OpenAI is developing reasoning technology under the code name “Strawberry,” formerly known as Q*. This project is seen as so groundbreaking that several researchers expressed concerns in a letter to OpenAI’s board in November, warning about its potential risks to humanity.
