OpenAI has changed the way that its new ChatGPT o3-mini model displays its chain of thought (Cot) to “make it easier for people to understand how the model thinks,” however it has faced an almost immediate backlash from users and accusations that it is copying the way that DeepSeek’s R1 model displays its reasoning.
Speaking to TechRadar, OpenAI said: “Users have told us that understanding how the model reasons through a response not only supports more informed decision-making but also helps build trust in its answers.”
“While the model’s raw CoT remains hidden as it’s hard to understand, we’ve found a balance: the model can think freely, and then it organises those thoughts so that they are easy to read. To improve clarity and safety, we’ve added an additional post-processing step where the model reviews the raw chain of thought, removing any unsafe content and then simplifies any complex ideas. Additionally, this post-processing step enables non-English users to receive the CoT in their native language, creating a more accessible and friendly experience.”
The new approach effectively provides summaries of the model’s reasoning instead of showing you the raw data. However, despite the new approach being available in both the o3-mini for free and paid users and o3-mini-high for paid users, the response on X to its changes was not entirely positive.
Mayo Oshin responded on X, posting: “We’d appreciate if you showed the full chain of thought, not just summarised version….thank you”, and Conor responded with, “its still a summary and not real CoT, which is disappointing.”
Some other users responded by suggesting that OpenAI was simply responding to the threat offered by the new DeepSeek by copying the way it presented the reasoning chain in its R1 model. “Finally DeepSeek changing the O-World for us,” replied Hamza. “So OpenAI copied Deepseek’s Chain of Thought feature?” said Ignis Rex, and “That moment when China is the one innovating, and US is the copycat,” said Josip Tomo Licardo.
A big step forward
Personally, I much prefer the new way of presenting reasoning in ChatGPT o3-mini. If you try the model now (just hit the Reasoning button) before you enter your prompt, you’ll find that much more information is provided on how ChatGPT is coming to its conclusion. The previous scarcity of reasoning information, when compared to DeepSeek, was something I criticized 03-mini for in the last few days. So, from my perspective, I think the new approach is a big step forward for ChatGPT o3-mini.
I’m not alone in welcoming the new feature, either. X user Roman Pshichenko said, “Hilariously, the chain of thought is more endearing and funny than the stiff response,” and “That’s a nice improvement! I’d love to see some examples of how this update impacts it’s reasoning process,” said Jesus Vazquezon X.
Leave a comment