OpenAI is preparing to launch a new AI model, internally named 'Strawberry,' which is designed to perform complex reasoning tasks similar to human capabilities. A limited release could occur as soon as this week, according to sources familiar with the situation.
This model represents a significant advancement in AI technology, allowing it to tackle multi-step problems, including complicated math and coding queries. The updated system will differ from the existing ChatGPT experience by incorporating a 'chain of thought' prompting technique, which pauses to evaluate related prompts before generating a response.
The new approach aims to improve accuracy in responses to challenging questions that have previously confounded AI systems. For example, the model has successfully addressed comparisons of numerical values, such as determining that 9.9 is greater than 9.11.
OpenAI has been focusing on enhancing its AI's reasoning abilities for some time, including a recent initiative to train models through multi-step problem-solving processes. This development comes amid increased competition in the AI sector from companies like Anthropic and Google, which are also advancing their own reasoning capabilities.
Noam Brown, a research scientist at OpenAI, is scheduled to discuss generative AI and multi-step reasoning at an upcoming TED AI event in San Francisco, highlighting the company's commitment to advancing AI technology.