In response to pressure from rivals, including Chinese AI company Deepseek, Openai is changing the way its latest AI model, O3-Mini, communicates a step-by-step “thinking” process.
Openai on Thursday announced an updated “Chain of Thought” showcasing how free and paid users of ChatGpt, an AI-powered Chatbot platform, have reached the model's “inference” steps and answers to questions It has been announced that it will be displayed. According to OpenAI, subscribers to the Premium ChatGPT plan that use O3-MINI with a “high inference” configuration will also see this updated read.
“We're introducing an update [chain of thought] An Openai spokesperson told TechCruch via email. “This update will help you follow the model's inference, make it more clear and give you confidence.”
Image credit: Openai
Inference models like O3-Mini are about thoroughly checking yourself before you produce results. This helps avoid some of the pitfalls that usually make the model trip. The trade-off is that it takes a little longer for the inference model to reach the solution – usually seconds to minutes longer.
Deepseek's R1 model, a “inference” model along the O3-Mini line, reveals its complete thought process. This is argued that many AI researchers prefer. In addition to making the model easier to study, the reasoning procedure provides a better user experience in certain situations and helps to show when the model is right or wrong.
Openai had chosen not to show the complete inference step of O3-Mini and its predecessors, O1 and O1-Mini, for competitive reasons. Instead, the user only looked at the outline of the inference steps – sometimes incorrect summary.
When we explained people before the release of O1-Preview, watching COT live was usually a “Aha” moment for them, who revealed that this was a big deal. These are not raw beds, but they are close to a big step and I am pleased to share that experience with the world. https://t.co/72zpprhmfk
– Gnom Brown (@polynoamial) February 6, 2025
Openai has yet to show the complete inference step for the O3-Mini, but the company says it has “find the balance.” The O3-Mini can “think freely” and organize its “thinking” into a more detailed summary.
“To improve clarity and safety, the model reviewed the raw set of ideas, removed unsafe content, and added additional post-processing steps to simplify complex ideas,” Openai spokesperson said. Mann continued. “In addition, this post-processing step allows non-English users to receive a series of thoughts in their native language, creating a more accessible and friendly experience.”
At Reddit AMA last week, Openai's Chief Product Officer Kevin Weil hinted at the coming change.
“We're committed to showing more than we can show today — [showing the model thought process] It will soon, it will soon,” he said. “All TBD – Showing a chain of thoughts leads to competitive distillation, but we also know that (at least the power users) want it, so the right way to balance it I'll find it.”
TechCrunch has a newsletter focused on AI! Sign up here to get it every Wednesday in your inbox.