ChatGPT Acquires New o1 Model, First to Offer “Reasoning” for Difficult Issues

OpenAI announced in a blog post on Thursday the introduction of a new model called o1, which is designed to tackle more complex problems, evaluate its responses, explore various strategies, and enhance its reasoning capabilities.

This model is currently available in two versions: o1-preview and o1-mini. It has achieved an impressive ranking in the 89th percentile in Codeforces’ competitive programming contests, is among the top 500 students in the United States for the Math Olympiad, and demonstrates “exceeds PhD-level accuracy on a benchmark of physics, biology, and chemistry problems,” as reported by OpenAI.

According to Jerry Tworek, OpenAI’s research lead, “We have noticed that this model hallucinates less,” in an interview with The Verge. The o1 model has been developed using a new optimization algorithm and a specially curated training dataset. Unlike previous models that focused on replicating patterns from their training data, o1 employs reinforcement learning, which facilitates its learning through a system of rewards and penalties.

A report from The Information on Tuesday highlights that the distinguishing feature of the o1 model compared to its predecessors is its capacity for “thoughtful” processing. This capability allows the model to take 10 to 20 seconds to formulate a considered response, rather than providing immediate answers. The o1 model, informally dubbed “Strawberry” by observers—likely about the viral trend of influencers querying AIs about the number of “Rs” in “strawberry”—eliminates the necessity for “chain-of-thought prompting.” This means users no longer need to pose additional questions to elicit the AI’s intermediate reasoning, as the model is inherently designed to present its reasoning process by default.

However, it is important to note that o1 is currently in its preview phase, which entails several significant limitations. Unlike GPT-4o, the o1 model lacks internet connectivity, does not support file uploads, and has various API restrictions for developers. In contrast, the o1-mini model is tailored to provide rapid responses for STEM-related inquiries.

Thank you for reading this post, don't forget to follow my whatsapp channel