On Thursday, Google part of Alphabet, took big steps in AI. They showed off two AI systems that can tackle tricky math problems. This breakthrough moves generative AI forward in areas that need deeper thinking – something regular AI has found tough.
Today’s AI models guess the next word by looking at data patterns. This works fine for language tasks, but abstract math needs more brain-like thinking. This has been a big hurdle for AI systems to overcome.
Google’s AI branch DeepMind just shared some cool findings from its new projects: AlphaProof and AlphaGeometry 2. They put these systems to the test with questions from the 2024 International Math Olympiad, a top-notch event for high schoolers. The AI models nailed four out of six tough questions setting a new AI record in this contest.
One question was cracked in mere minutes, but others took up to three days to solve—longer than the competition allows. Even so, Google pointed out that these results show the best AI performance ever seen in the Math Olympiad’s history.
AlphaProof, a system that showed remarkable reasoning skills, came about by joining the Gemini language model (the core of Google’s chatbot) with AlphaZero, another AI system that beat humans at games like chess and Go. This combo enabled AlphaProof to crack three of the Olympiad’s questions, including the toughest one, which five out of over 600 human participants managed to solve.
These advances follow similar breakthroughs in AI. In early July, Reuters said Microsoft-backed OpenAI was also developing AI tech for reasoning called “Strawberry.” This project once named Q*, has grabbed attention because it could cause a revolution in AI, but it’s not without debate. Some OpenAI researchers even raised red flags telling the company’s board that the tech might have unforeseen risky effects on humanity.
As AI keeps making quick strides in fields that need deep thinking, the outlook for artificial intelligence looks bright yet tricky. While firms such as Google keep pushing what AI can do, people still argue a lot about how these systems will affect things from schooling to moral issues.
