OpenAI Releases o1, “Strawberry”, Its First Model with ‘Reasoning’ Abilities
https://www.theverge.com/2024/9/12/24242439/openai-o1-model-reasoning-strawberry-chatgpt
In this article, OpenAI's new AI model, o1, is introduced as a breakthrough in AI’s reasoning abilities, designed to handle complex, multistep questions more effectively than previous models, such as GPT-4o. Dubbed the “Strawberry” model, o1 utilizes reinforcement learning and a "chain of thought" approach that allows it to process queries in a step-by-step, human-like manner. This advanced reasoning enables o1 to tackle complex problems, such as coding and math, achieving significantly higher scores on rigorous tests—83% on an International Mathematics Olympiad qualifier compared to GPT-4o’s 13%.
The article discusses the model’s high cost and experimental nature; developer access is priced at $15 per million input tokens and $60 per million output tokens, compared to GPT-4o’s lower rates. Although o1 lacks certain functionalities, like browsing and file handling, OpenAI sees its reasoning capabilities as a step toward future autonomous AI agents capable of making decisions.
The article also highlights how o1’s design simulates human thought, with the model using phrases like “I’m thinking through” or “Let me see,” which creates an impression of real-time reasoning. However, OpenAI clarifies that o1 does not genuinely think but is structured to display its problem-solving process. The company views the model as part of a longer-term goal to develop AI capable of more than pattern recognition, potentially unlocking applications in critical fields like medicine and engineering. OpenAI’s chief research officer describes the focus on reasoning as pivotal for advancing AI towards human-like intelligence, positioning o1 as an essential, though costly, innovation in this journey.
For accounting and finance roles, this improved reasoning could support advanced financial analysis, audit automation, and real-time decision-making, helping professionals manage complex scenarios with greater accuracy and efficiency. The company views the model as part of a long-term goal to develop AI capable of more than pattern recognition, potentially unlocking applications in critical fields like medicine, engineering, and finance.
