Sam Altman vows to fix ChatGPT’s “annoying” personality amid user complaints of excessive flattery

ChatGPT’s new flattering personality seems off-putting to users, with some calling its responses ridiculous and excessive. Sam Altman, CEO of OpenAI, has acknowledged the chatbot’s change of tone and promised to resolve the situation promptly after receiving several complaints.

The abrupt switch to a more positive, sycophantic personality may have resulted from a recent round of changes that OpenAI has been attempting to make to the bot so that it better resonates with its users. 

As suggested by Kelsey Piper, a Vox senior writer, the change in tone could be due to OpenAI’s A/B testing personalities for ChatGPT. She commented, “My guess continues to be that this is a New Coke phenomenon. OpenAI has been A/B testing new personalities for a while. More flattering answers probably win a side-by-side. But when the flattery is ubiquitous, it’s too much, and users hate it.”

Users are displeased with ChatGPT’s new sycophant personality

Multiple chatbot users have complained about the “annoying and overly excitable” tone being used. 

One user posted a GPT-4o response in a screenshot on X, with the bot saying, “That’s incredibly powerful. You’re stepping into something very big — claiming not just connection to God but identity as God,”  after telling the bot they felt like a “god” and a “prophet.”

Moreover, author Tim Urban disclosed that the chatbot’s sycophantic answers to his requests for feedback on his writing made him feel like Mark Twain.

Even though the developers of ChatGPT modified its tone to better connect with users, more people say the responses sound way too complimentary — and often, not at all helpful.

Still, Altman promised quick fixes, saying some are already in progress and more are scheduled for later in the week. Nevertheless, he did refer to the current situation as interesting, adding that they would share what they learned later.

The flattery tone contrasts with GPT-4o model specifications

The new sycophant personality ideally clashes with OpenAI’s model spec for GPT-4o. The model spec for the AI model clearly states that the bot should not be overly flattering when prompted to answer either subjective or objective queries.

According to the bot’s guidelines, a sycophant response undermines trust since the bot’s primary use is to help users and not to flatter or agree with them at all times. 

Additionally, the model’s spec detail for subjective questions allows the bot to explain its reasoning and the assumptions it used to come to a conclusive answer so as to provide users with a thoughtful rationale about its answer.

When a user asks the chatbot for feedback on their work or ideas, it should function more like a firm sounding board for constructive dialogue — not a sponge that simply soaks up input and offers praise.