OpenAI has announced a new artificial intelligence system, OpenAI o3, designed to reason through complex problems in math, science, and computer programming. The system outperformed industry-leading A.I. technologies on standardized benchmark tests, showcasing its accuracy in math, science, coding, and logic tasks.
o3 is the successor to o1, introduced earlier this year, with improved accuracy of over 20% in common programming tasks. It also surpassed OpenAI’s chief scientist, Jakub Pachocki, on a competitive programming test. The company plans to roll out the technology to individuals and businesses early next year.
OpenAI CEO Sam Altman described o3 as “incredible at programming.” However, he noted that one skilled programmer could still beat the system in certain tests. This new technology aims to build A.I. systems that can carefully solve complex problems through logical steps.
The development of o3 follows similar efforts by Google with its Gemini 2.0 Flash Thinking Experimental. These technologies aim to create A.I. systems capable of logical problem-solving, useful for computer programmers and students seeking assistance from automated tutors in math and science.
OpenAI’s initial success with ChatGPT demonstrated the potential for machines to handle requests like humans, answering questions, writing term papers, and generating code. However, the responses were sometimes flawed due to learning patterns from untrustworthy internet sources. The new system employs reinforcement learning, a process that identifies patterns through extensive trial and error.
While o3 is designed to reason, it may still make mistakes or hallucinate like its predecessors. Its computing requirements can be substantial, making it expensive. OpenAI has already begun selling o1 to professionals for $200 per month, and the new o3 system will follow suit when released early next year.
Source: https://www.nytimes.com/2024/12/20/technology/openai-new-ai-math-science.html