ChatGPT gets new o1 model, first with ‘reasoning’ for difficult problems
ChatGPT has a new model called o1 that has been trained to solve more difficult problems, analyze the answers, try different strategies and refine its thinking, OpenAI said in a blog post on Thursday.
The new model, currently split between o1-preview and o1-mini, ranks in the 89th percentile of Codeforces’ competitive programming contests, is among the top 500 students in the US in the Math Olympiad, and “exceeds PhD-level accuracy on a benchmark of physics, biology, and chemistry problems,” according to OpenAI.
“We found that this model hallucinates less,” said Jerry Tworek, research lead at OpenAI, in an interview with The Edge. It is trained on a novel optimization algorithm with a custom training dataset. While previous models have relied on mimicking patterns in their training data, o1 uses reinforcement learning, which teaches it via rewards and punishments.
What sets o1 apart from previous models is its ability to “think,” according to a report by The information on Tuesday. This means that the model doesn’t start spitting out responses right away and can take 10 to 20 seconds to formulate a thoughtful response. The o1 model, which is also nicknamed “Strawberry” by onlookers (a possible reference to the viral trend of Influencers Asking AIs How Many “R’s” Are in the Word “Strawberry”), eliminates the need for “chain-of-thought prompting,” where users have to ask an AI additional questions to see its intermediate reasoning. Instead, the model is designed to show its reasoning by default.
Because o1 is still in preview, there are a number of significant limitations. Unlike GPT-4o, o1 is not web-connected, cannot be used with file uploads, and has a multitude of API limitations for developers. The o1-mini model differs in that it focuses on providing quick answers to STEM-related questions.
The competition in the AI space is heating up, as every player in the big tech space tries to outdo each other and create “agentive” AIs that can complete tasks for you. Earlier this year at Google I/O, the search giant unveiled a more powerful version of Gemini that can converse with you more naturally and even let you interrupt it mid-sentence. And at its iPhone 16 launch event earlier this week, Apple ramped up the processing power of its latest phones to handle Apple Intelligence, a suite of AI features coming to iPhones powered by its OpenAI technology.
While the AI hype has driven tech stocks to record numbers over the past two years, it appears investors are becoming more cautious. Nvidia, the chipmaker that creates the brains that power many of the world’s best AI data centers, saw a 10% drop last weekThe tech world in general may be cooling off on AI as it waits for more concrete results from services, though that hasn’t stopped OpenAI from making a stunning Valuation of $150 billion.
For ChatGPT Plus and Team users, the o1 preview model is rolling out now. ChatGPT Enterprise and Edu users will get access next week. Developers can also use the API for prototyping.
