Although Google claims to have unveiled a new “reasoning” AI model, it is still in the experimental phase, and based on our quick testing, there is undoubtedly space for improvement.
Google’s AI prototype platform, AI Studio, offers the new model, which is aptly named Gemini 2.0 Flash Thinking Experimental. It is “best for multimodal understanding, reasoning, and coding” and can “reason over the most complex problems” in areas like programming, physics, and math, according to a model card.
Logan Kilpatrick, the product lead for AI Studio, referred to the Gemini 2.0 Flash Thinking Experimental as “the first step in [Google’s] reasoning journey” in a post on X. According to Jeff Dean’s blog post, the Gemini 2.0 Flash Thinking Experimental is “trained to use thoughts to strengthen its reasoning.” Dean is the head scientist for Google DeepMind, the company’s AI research branch.
In reference to the amount of computing required to “run” the model while it evaluates a query, Dean stated, “We see promising results when we increase inference time computation.”
The Gemini 2.0 Flash Thinking Experimental, which is based on Google’s recently revealed Gemini 2.0 Flash model, looks to share architectural similarities with OpenAI’s o1 and other so-called reasoning models. Reasoning models, in contrast to most AI, are able to fact-check themselves, which helps them avoid some of the common mistakes that AI models make.
One disadvantage of reasoning models is that they frequently take longer to get answers, typically ranging from seconds to minutes.
The Gemini 2.0 Flash Thinking Experimental pauses before answering a prompt, taking into account several related prompts and “explaining” its reasoning as it goes. The model eventually summarizes what it believes to be the most correct response.
That’s what’s meant to occur, though. Gemini 2.0 Flash Thinking Experimental responded, “two,” when I asked how many R’s were in the word “strawberry.”
Read More News
