Google develops “reasoning” AI to rival OpenAI

Google (GOOGLE) would like to take its artificial intelligence rivalry with OpenAI to the next level of AI models: those that can “reason”.

Google teams have made progress in recent months on software for AI models that resemble human reasoning abilities, Bloomberg reported, citing unnamed people familiar with the matter. Google has long focused on “reasoning” abilities in large language models (LLM), particularly in his work on chain of thought prompt.

With this technique, which Google uses according to Bloomberg, LLMs can solve multi-step problems using “a series of intermediate reasoning steps,” the same way a human would. Models with this software take longer to respond to requests because they consider similar prompts summarized in a response. The chain of thought technique gives models the ability to solve more complex questions related to mathematics and computer programming.

OpenAI is also using chain of thought prompts for his new model o1. After the model, internally called Strawberry, was released in September, some employees at Google’s DeepMind unit worried the company was falling behind, one person told Bloomberg. However, they added that employees are no longer worried now that Google has revealed more competitors to OpenAI products.

Unlike the current version of ChatGPT, OpenAI new model doesn’t yet have some of its “useful” features, such as browsing the web and downloading files and images. The o1 Series models are “designed to spend more time thinking before responding,” the company said. The models, the first of which are available to preview in ChatGPT and through the company’s API, can “reason” about more complex tasks and problems in science, coding, and math than previous OpenAI models.

Google is also working to improve the “reasoning” capabilities of its Gemini chatbot. In July, Google made its fastest and most cost-effective model, 1.5 Flash, available in the non-paid version of Gemini. With Flash 1.5, Gemini will have “faster, more useful responses,” Google said, adding that users will notice improvements in Gemini’s reasoning and image processing capabilities.