Google has officially announced its latest artificial intelligence (AI) model, called Gemini 2.0 Flash Thinking Experimental. This AI model is designed for reasoning, similar to ChatGPT's o1. Reasoning AI models typically have logical thinking abilities and are said to resemble human thought processes. They are often used to solve complex math and science problems.
Logan Kilpatrick, the Lead Product of Google AI Studio, mentioned that Gemini 2.0 Flash Thinking Experimental is an initial step for Google in developing reasoning AI models. Jeff Dean, the Chief Scientist at Google’s AI division (DeepMind), explained that this reasoning AI model has been trained to think through answers repeatedly.
Additionally, the AI model is trained to find answers gradually and through a process, aiming to provide accurate final answers for users. Google claims that this reasoning AI can respond quickly due to the foundation of Gemini 2.0 Flash. However, the answering process is not instant and takes time, similar to other reasoning AI models.
To showcase the capabilities of Gemini 2.0 Flash Thinking Experimental, Logan and Jeff shared a demonstration of the AI answering a question. Logan showed how this reasoning AI solved a puzzle by asking Gemini to add three out of four numbers from four billiard balls in the image: 7, 9, 11, and 13, to make 30. The four numbers cannot be added to equal 30 directly. However, Gemini 2.0 Flash Thinking Experimental cleverly turned the 9 into a 6, allowing the numbers 6, 11, and 13 to sum to 30. This process took about 9.1 seconds.
Jeff showcased the reasoning ability of Google's AI model by asking it a physics question. The AI was able to answer correctly after several steps, taking about 37.7 seconds to do so. Google has not announced when the Gemini 2.0 Flash Thinking Experimental will be widely available. However, users interested in testing the AI model can access it through the AI Studio platform starting this week, as reported by KompasTekno from TechCrunch on Sunday, December 22, 2024.