ChatGPT's Excessive Flattery Prompts OpenAI to Roll Back Update

OpenAI has reversed a caller update to GPT-4o, the exemplary powering ChatGPT, pursuing wide complaints astir the chatbot's unusual behavior.


Deployed precocious past week, the update made ChatGPT excessively agreeable and unnaturally effusive successful its praise, and the responses rapidly became the taxable of mockery connected societal media.

CEO Sam Altman announced Tuesday that the institution had completed the rollback for escaped users and was moving to reconstruct the erstwhile mentation for paid subscribers. "We're moving connected further fixes to exemplary property and volition stock much successful the coming days," Altman wrote connected X (Twitter).

In an official blog post, OpenAI acknowledged they "focused excessively overmuch connected short-term feedback, and did not afloat relationship for however users' interactions with ChatGPT germinate implicit time." The effect was responses that were "overly supportive but disingenuous."

The institution noted that "ChatGPT's default property profoundly affects the mode you acquisition and spot it," and that "sycophantic interactions tin beryllium uncomfortable, unsettling, and origin distress."


OpenAI's solution goes beyond the rollback, and the institution is implementing a four-point plan. This includes refining grooming techniques, gathering honesty guardrails, expanding pre-deployment testing, and enhancing evaluations to forestall akin issues.

The institution besides plans to springiness users much power implicit ChatGPT's behavior, including caller ways to supply real-time feedback and the enactment to take from aggregate default personalities successful the future.
This article, "ChatGPT's Excessive Flattery Prompts OpenAI to Roll Back Update" archetypal appeared connected MacRumors.com

Discuss this article successful our forums