OpenAI’s new GPT-5.3 Instant: Less “cringe” tone, no more “over-caveating” responses
OpenAI has released GPT-5.3 Instant, a conversational model that the company says will have smoother, more-to-the-point conversations.
It seems OpenAI is walking back from its cautious guardrails, allowing a more permissive AI chatbot.
In a video describing the new model, a researcher explained, “People are noticing that our models can sometimes seem like a bit of a nanny.” The company describes this over-cautious behavior as “over-caveating,” and this new model aims to relax a bit and let more things slide.
An example showed a question about calculations for “a really long-distance archery scenario.” The previous version of the model included caveats that the AI could not help with calculations that could be harmful. The new model’s response just went ahead and answered the question, without assuming any bad intent.
The researcher in the video said that these changes did not loosen safety controls, but rather just improved contextual understanding of the user’s query.
The model also features more useful web searches, which attempt to infer the context of why the user is asking the question, and then tailor a more useful response. The company says results won’t just appear to be a list of links, but rather a more direct response with the information the user was looking for.
The company said the new model is available today for all users of ChatGPT.
In a video describing the new model, a researcher explained, “People are noticing that our models can sometimes seem like a bit of a nanny.” The company describes this over-cautious behavior as “over-caveating,” and this new model aims to relax a bit and let more things slide.
An example showed a question about calculations for “a really long-distance archery scenario.” The previous version of the model included caveats that the AI could not help with calculations that could be harmful. The new model’s response just went ahead and answered the question, without assuming any bad intent.
The researcher in the video said that these changes did not loosen safety controls, but rather just improved contextual understanding of the user’s query.
The model also features more useful web searches, which attempt to infer the context of why the user is asking the question, and then tailor a more useful response. The company says results won’t just appear to be a list of links, but rather a more direct response with the information the user was looking for.
The company said the new model is available today for all users of ChatGPT.
