ChatGPT update adds canvas sharing and improved reasoning transparency

ChatGPT recently released a series of improvements. First of all, a major update relates to canvas sharing. Users can now share their canvases by generating a public URL. When someone opens this link, they will see a rendered version of the canvas. Additionally, there is a button that allows them to edit the canvas within their own ChatGPT session. This feature is similar to Artifact Remix on Claude, enabling easier collaboration and content sharing.

Besides that, OpenAI has also announced improvements to the chain of thought process in O3 Mini. The goal of this update is to make the reasoning process more transparent. Previously, ChatGPT provided generic summaries, but now it aims to reflect the actual thought process in a clearer way. However, users have noticed that the output still feels more like a summary rather than a true step-by-step reasoning breakdown. OpenAI has acknowledged this limitation.

Later, it was discovered that the custom system prompt for this feature had been hardcoded within the experimental configuration. This made it possible to extract the full system prompt that dictates how the chain of thought summarizer should behave. The extracted prompt contained detailed instructions for ChatGPT, explaining how to write these summaries, what to include, and what to avoid. It also featured multiple examples to guide the model’s behavior.

This prompt appears to be a well-structured guide for reasoning models in general. Researchers and AI enthusiasts may find it valuable for studying how such prompts influence model behavior. If you’re interested, you can also take a look at this system prompt yourself.