OpenAI CEO Sam Altman discusses the potential of the company’s new o1 model

OpenAI CEO Sam Altman recently unveiled the company’s new o1 model, codenamed Strawberry, at a T-Mobile event in San Francisco. Altman compared the current state of o1 to where OpenAI’s language models were when GPT-2 was released in 2019, hinting at significant improvements in the coming years similar to the progression from GPT-2 to GPT-4.

Unlike most generative AI models, o1 is unique in its ability to plan out its approach when responding to queries and explore multiple approaches before providing an answer. This feature sets it apart from other models like GPT-4o, which begin answering immediately without exploring different options.

OpenAI has already rolled out a preview of o1 and a smaller model called o1 mini, specifically designed for coding. The o1 model is currently most valuable in solving problems in math, science, and coding, with users already creating unusual and unexpected projects with it.

The tech industry is rapidly advancing AI capabilities, with OpenAI’s reasoning advancements representing one path towards more sophisticated AI systems. Altman highlighted that o1 is transitioning OpenAI from the first level of AI capabilities to the beginning of the second level, where AI can achieve human-level problem-solving.

However, OpenAI has rated o1 as a “medium risk” on its safety scorecard, citing concerns about the AI’s persuasive abilities and potential risks related to weapon development. Despite its capabilities, the company emphasizes the need to lower expectations for o1 in the short term, acknowledging that the model is still flawed and limited.

Overall, OpenAI’s o1 model shows promise for the future of AI technology, with ongoing improvements expected to enhance its capabilities and impact across various industries.