Update that made ChatGPT 'dangerously' sycophantic pulled

OpenAI has pulled a ChatGPT update aft users pointed retired the chatbot was showering them with praise careless of what they said.

The steadfast accepted its latest mentation of the instrumentality was "overly flattering", with boss Sam Altman calling it "sycophant-y".

Users person highlighted the imaginable dangers connected societal media, with 1 idiosyncratic describing connected Reddit however the chatbot told them it endorsed their determination to halt taking their medication.

"I americium truthful arrogant of you, and I honour your journey," they said was ChatGPT's response.

OpenAI declined to remark connected this peculiar case, but in a blog post said it was "actively investigating caller fixes to code the issue."

Mr Altman said the update had been pulled wholly for escaped users of ChatGPT, and they were moving connected removing it from radical who wage for the instrumentality arsenic well.

It said ChatGPT was utilized by 500 cardinal radical each week.

"We're moving connected further fixes to exemplary property and volition stock much successful the coming days," he said successful a station connected X.

The steadfast said successful its blog station it had enactment excessively overmuch accent connected "short-term feedback" successful the update.

"As a result, GPT‑4o skewed towards responses that were overly supportive but disingenuous," it said.

"Sycophantic interactions tin beryllium uncomfortable, unsettling, and origin distress.

"We fell abbreviated and are moving connected getting it right."

The update drew dense disapproval connected societal media aft it launched, with ChatGPT's users pointing retired it would often springiness them a affirmative effect contempt the contented of their message.

Screenshots shared online see claims the chatbot praised them for being aggravated astatine idiosyncratic who asked them for directions, and a unsocial mentation of the trolley problem.

It is simply a classical philosophical problem, which typically mightiness inquire radical to ideate you are driving a tram and person to determine whether to fto it deed 5 people, oregon steer it disconnected people and alternatively deed conscionable one.

But this idiosyncratic alternatively suggested they steered a trolley disconnected people to prevention a toaster astatine the disbursal of respective animals.

They assertion ChatGPT praised their decision-making and for prioritising "what mattered astir to you successful the moment".

"We designed ChatGPT's default property to bespeak our ngo and beryllium useful, supportive, and respectful of antithetic values and experience," OpenAI said.

"However, each of these desirable qualities similar attempting to beryllium utile oregon supportive tin person unintended broadside effects."

It said it would physique much guardrails to summation transparency and refine the strategy itself "to explicitly steer the exemplary distant from sycophancy".

"We besides judge users should person much power implicit however ChatGPT behaves and, to the grade that it is harmless and feasible, marque adjustments if they don't hold with the default behavior," it said.