OpenAI has released a new series of AI models called OpenAI o1 that are designed to enhance reasoning capabilities for solving complex problems. The o1-preview and o1-mini models aim to give machines more time to think through a problem before generating a response, which could be useful in fields like science, coding, and math.
OpenAI reports that these models learn to refine their thought processes through training, allowing them to try different strategies and recognize mistakes. In tests, the upcoming model updates performed on par with PhD students on difficult benchmark physics, chemistry, and biology problems. The inference model significantly outperformed previous models, solving 83% of the problems in the International Mathematical Olympiad qualifying exams, compared to 13% for GPT-4.
For developers, the o1 series offers enhanced coding capabilities, reaching the 89th percentile in Codeforces competitions. The OpenAI o1-mini is a smaller, more cost-effective model that is 80% cheaper than the o1-preview and excels at generating and debugging complex code.
These advances could have an impact on the crypto industry, where complex code and mathematical reasoning are key. The improved reasoning and coding capabilities of the o1 model could be useful for developing smart contracts, analyzing blockchain protocols, and security audits.
OpenAI also implemented a new safety training approach to these models, inferring policy through thought chains to better adhere to safety and alignment guidelines. In challenging jailbreak tests, the o1-preview model scored significantly higher in maintaining adherence to safety rules compared to GPT-4.
Greg BrockmanThe president and co-founder of OpenAI said the O1 technology offers new security opportunities, demonstrating improvements in reliability, hallucination and robustness against adversarial attacks. The model's ability to reason step-by-step unlocks “System II thinking” to handle more complex tasks.
The o1 model is currently available to ChatGPT Plus and Team users, and will also be accessible to Enterprise and Edu users. Developers with qualifying API usage tiers can start prototyping on both models, although certain features such as function calls and streaming are not yet supported.
OpenAI will continue to develop and release the GPT and o1 series models, aiming to make them more useful by adding features such as browsing and uploading files and images.
Mentioned in this article
