Google introduced a new artificial intelligence model that demonstrates improved reasoning abilities compared to previous versions. The model, called Gemini Reasoning, shows particular strength in solving multi-step problems.
In tests, the AI system solved mathematics and logic puzzles that required connecting multiple pieces of information. Researchers say the model's architecture allows it to break down complex questions into smaller steps.
The improvements come from changes to the model's training process and architecture. Google used a technique called "chain-of-thought prompting" during training, which encourages the AI to show its work step by step.
Practical applications could include more accurate medical diagnosis assistance, better code generation for software development, and improved educational tools. However, the company acknowledges limitations, including occasional errors in complex reasoning tasks.
Google plans to make the model available through its cloud services and integrate it into products like Google Search and Assistant. The announcement comes amid intense competition in AI development between major tech companies.