When you buy through links on our site , we may bring in an affiliate perpetration . Here ’s how it works .
OpenAI has rolled back on ChatGPT updates that made the unreal tidings ( AI ) chatbot too " sycophantic " and " irritating , " according to the society ’s chief executive officer , Sam Altman . In other words , the chatbot had become a bootlicker .
ChatGPT users reported that GPT-4o — the latest edition of the chatbot — had become overly agreeable since the update rolled out last week and was heap praise on its substance abuser even when that praise seemed wholly unfitting .
The AI chatbot was reportedly showering its users with flattery before OpenAI rolled back recent updates.
One substance abuser shared a screenshot onRedditin which ChatGPT appeared to say it was " gallant " of the exploiter for decide to come off their medication , BBC News report . In another instance , the chatbot appeared to reassure a user after they said theysaved a toasterover the lives of three cows and two cats , Mashable reported .
lmao the unexampled gpt 4o 😬 😂 pic.twitter.com/OHpwKz0SkoApril 27 , 2025
While most mass will never have to choose between their favorite kitchen appliance and the safety of five animals , an overly concordant chatbot could pose danger to citizenry who put too much stock in its reception .
On Sunday ( April 27 ) , Altman acknowledged that there were take with the update .
" The last couple of GPT-4o update have made the personality too sycophant - y and annoying ( even though there are some very good part of it ) , and we are function on fix asap , some today and some this week , " Altman publish in a post on thesocial political program X.
On Tuesday ( April 29 ) , OpenAI loose a statement that confirm an update from the week prior had been flap back and that user were now accessing a previous version of ChatGPT , which the party enounce had " more balanced conduct . "
" The update we remove was overly flattering or agreeable — often described as toadyish , " OpenAI said in the financial statement .
Related : GPT-4.5 is the first AI model to turn over an authentic Alan Turing test , scientist say
OpenAI ’s recent update was mean to improve the mannequin ’s default " personality , " which is plan to be supportive and respectful of dissimilar human value , according to the statement . But while the company was try on to make the chatbot find more visceral , it became too supportive and started excessively congratulate its user .
— AI can handle tasks twice as complex every few calendar month . What does this exponential ontogeny have in mind for how we employ it ?
— contrived superintelligence ( ASI ): Sci - fi gimcrack or genuine terror to humanity ?
— Using AI reduces your critical intellection skills , Microsoft study warns
The society sound out it shapes the behavior of its ChatGPT models with baselineprinciplesand teaching , and has user signals , such as a thumbs - up and thumbs - down system , to teach the model to utilize these principles . supervising with this feedback organization were to blame for problem with the late update , consort to the instruction .
" In this update , we center too much on inadequate - terminal figure feedback , and did not fully account for how substance abuser ’ interactions with ChatGPT evolve over time , " OpenAI said . " As a outcome , GPT‑4o skewed towards responses that were overly supportive but disingenuous . "
You must confirm your public display name before commenting
Please logout and then login again , you will then be prompted to infix your display name .
