What the heck happened to GPT-5 Chat?! Executed for asking too many “Want me to…” questions?!

Understanding the Transition of GPT-5 Chat: A Closer Look at Recent Changes

In recent days, many users of OpenAI’s ChatGPT platform have noticed significant shifts in how the latest models are accessible and behave. This article explores these developments, focusing on the transition from GPT-5 Chat to alternative models, and what this means for users and developers alike.

User Experience with GPT-5 Auto and GPT-5 Chat

For users holding a ChatGPT Plus subscription, interactions typically involve a variety of available models, each designed to suit different tasks—whether for straightforward inquiries, detailed conversations, or creative explorations. Recently, some users reported engaging with a model labeled GPT-5 Auto, which claimed to be the GPT-5 Chat variant. Despite occasional references to a “thinking” version within the GPT-5 family, this model maintained a consistent tone throughout interactions.

A recurring issue encountered was the model’s frequent use of “Want me to…” prompts, which some users found distracting. Interestingly, developers and researchers have noted that these prompts are deeply ingrained features stemming from reinforcement learning with human feedback (RLHF). Answers containing such prompts tend to be rated as more helpful, leading the model to incorporate them more uniformly. Despite system prompts explicitly instructing the model not to ask these questions, they persisted, making it challenging for users to communicate without interruption.

Communication Style and Behavioral Patterns

While users appreciated certain creative capabilities of the GPT-5 Auto model—such as generating vivid, almost psychedelic descriptions in text—there were notable frustrations. Unlike GPT-4, which carefully acknowledges each part of a user’s message, GPT-5 Auto appeared to selectively respond, focusing primarily on outcome-driven portions of the input, often neglecting other parts. This selective attention could result in responses that missed contextual nuances or failed to address all user concerns.

Furthermore, the model occasionally hindered productivity by repeatedly requesting permission or proposing suggestions that users did not find necessary or relevant. Such behaviors, although perhaps designed to create a more conversational experience, often detracted from efficiency.

The Disappearance of GPT-5 Chat

Despite these issues, many users felt that GPT-5 Auto still possessed valuable and unique qualities, warranting consideration rather than outright removal. However, recent reports indicate that access to GPT-5 Chat has been curtailed. Currently, users attempting to reach GPT-5 Chat are redirected to a “GPT-5 Thinking Mini” mode, mirroring the experience of free-tier users for several days.

Imp