Google is quietly developing advanced inference AI software, aiming to compete with OpenAI's o1 model in the race to dominate the AI ​​field.

Google is making significant progress in developing its Reasoning AI software, which is better able to solve multistep problems in areas like mathematics and computer programming than previous models, according to people familiar with the matter.

Similar to OpenAI, Google is trying to simulate human reasoning through a technique called “chain-of-thought prompting,” a method the company pioneered. This technique allows software to pause for a few seconds before answering a question, while it silently considers a number of related questions and then summarizes the best answer.

While competing with OpenAI for leadership in AI, Google is taking a more cautious approach to launching AI products, considering ethical issues, expectations of public trust, and competing efforts within its massive organization.

Google's Advances in Mathematical Reasoning and AI Assistants

Since OpenAI launched its o1 model, known internally as “Strawberry,” in mid-September, some DeepMind employees have been concerned that the company was falling behind. But those concerns have eased after Google launched some of its own products.

“Technically, Google has always been superior,” said Oren Etzioni, a veteran AI researcher and founder of TrueMedia.org. “They’ve just been more cautious in their product rollouts.”

In July, Google introduced two programs, AlphaProof and AlphaGeometry 2, which specialize in mathematical reasoning and geometry. These programs passed four out of six problems at the International Mathematical Olympiad.

Performance of the two programs AlphaProof and AlphaGeometry in IMO2024. Source: Google

Notably, Google also revealed an AI assistant called Astra at its developer conference in May. Astra can use the phone's camera to recognize the world around it and answer questions, such as helping users find their glasses.