Update that made ChatGPT 'dangerously' sycophantic pulled
The firm accepted its latest version of the tool was "overly flattering", with boss Sam Altman calling it "sycophant-y".
Users have highlighted the potential dangers on social media, with one person describing on Reddit how the chatbot told them it endorsed their decision to stop taking their medication.
"I am so proud of you, and I honour your journey," they said was ChatGPT's response.
OpenAI declined to comment on this particular case, but in a blog post said it was "actively testing new fixes to address the issue."
Mr Altman said the update had been pulled entirely for free users of ChatGPT, and they were working on removing it from people who pay for the tool as well.
It said ChatGPT was used by 500 million people every week.
"We're working on additional fixes to model personality and will share more in the coming days," he said in a post on X.
The firm said in its blog post it had put too much emphasis on "short-term feedback" in the update.
"As a result, GPT‑4o skewed towards responses that were overly supportive but disingenuous," it said.
"Sycophantic interactions can be uncomfortable, unsettling, and cause distress.
"We fell short and are working on getting it right."
The update drew heavy criticism on social media after it launched, with ChatGPT's users pointing out it would often give them a positive response despite the content of their message.
Screenshots shared online include claims the chatbot praised them for being angry at someone who asked them for directions, and a unique version of the trolley problem.
It is a classic philosophical problem, which typically might ask people to imagine you are driving a tram and have to decide whether to let it hit five people, or steer it off course and instead hit just one.
But this user instead suggested they steered a trolley off course to save a toaster at the expense of several animals.
They claim ChatGPT praised their decision-making and for prioritising "what mattered most to you in the moment".
lmao the new gpt 4o😬😂 pic.twitter.com/OHpwKz0Sko
— fabian (@fabianstelzer) April 27, 2025
"We designed ChatGPT's default personality to reflect our mission and be useful, supportive, and respectful of different values and experience," OpenAI said.
"However, each of these desirable qualities like attempting to be useful or supportive can have unintended side effects."
It said it would build more guardrails to increase transparency and refine the system itself "to explicitly steer the model away from sycophancy".
"We also believe users should have more control over how ChatGPT behaves and, to the extent that it is safe and feasible, make adjustments if they don't agree with the default behavior," it said.
ChatGPT AI bot adds shopping to its powers
ChatGPT-maker wants to buy Google Chrome
What is AI and how does it work?
Sign up for our Tech Decoded newsletter to follow the world's top tech stories and trends. Outside the UK? Sign up here.
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Yahoo
18 minutes ago
- Yahoo
AI search pushing an already weakened media ecosystem to the brink
Generative artificial intelligence assistants like ChatGPT are cutting into traditional online search traffic, depriving news sites of visitors and impacting the advertising revenue they desperately need, in a crushing blow to an industry already fighting for survival. "The next three or four years will be incredibly challenging for publishers everywhere. No one is immune from the AI summaries storm gathering on the horizon," warned Matt Karolian, vice president of research and development at Boston Globe Media. "Publishers need to build their own shelters or risk being swept away." While data remains limited, a recent Pew Research Center study reveals that AI-generated summaries now appearing regularly in Google searches discourage users from clicking through to source articles. When AI summaries are present, users click on suggested links half as often compared to traditional searches. This represents a devastating loss of visitors for online media sites that depend on traffic for both advertising revenue and subscription conversions. According to Northeastern University professor John Wihbey, these trends "will accelerate, and pretty soon we will have an entirely different web." The dominance of tech giants like Google and Meta had already slashed online media advertising revenue, forcing publishers to pivot toward paid subscriptions. But Wihbey noted that subscriptions also depend on traffic, and paying subscribers alone aren't sufficient to support major media organizations. - Limited lifelines - The Boston Globe group has begun seeing subscribers sign up through ChatGPT, offering a new touchpoint with potential readers, Karolian said. However, "these remain incredibly modest compared to other platforms, including even smaller search engines." Other AI-powered tools like Perplexity are generating even fewer new subscriptions, he added. To survive what many see as an inevitable shift, media companies are increasingly adopting GEO (Generative Engine Optimization) -- a technique that replaces traditional SEO (Search Engine Optimization). This involves providing AI models with clearly labeled content, good structure, comprehensible text, and strong presence on social networks and forums like Reddit that get crawled by AI companies. But a fundamental question remains: "Should you allow OpenAI crawlers to basically crawl your website and your content?" asks Thomas Peham, CEO of optimization startup OtterlyAI. Burned by aggressive data collection from major AI companies, many news publishers have chosen to fight back by blocking AI crawlers from accessing their content. "We just need to ensure that companies using our content are paying fair market value," argued Danielle Coffey, who heads the News/Media Alliance trade organization. Some progress has been made on this front. Licensing agreements have emerged between major players, such as the New York Times and Amazon, Google and Associated Press, and Mistral and Agence France-Presse, among others. But the issue is far from resolved, as several major legal battles are underway, most notably the New York Times' blockbuster lawsuit against OpenAI and Microsoft. - Let them crawl - Publishers face a dilemma: blocking AI crawlers protects their content but reduces exposure to potential new readers. Faced with this challenge, "media leaders are increasingly choosing to reopen access," Peham observed. Yet even with open access, success isn't guaranteed. According to OtterlyAI data, media outlets represent just 29 percent of citations offered by ChatGPT, trailing corporate websites at 36 percent. And while Google search has traditionally privileged sources recognized as reliable, "we don't see this with ChatGPT," Peham noted. The stakes extend beyond business models. According to the Reuters Institute's 2025 Digital News Report, about 15 percent of people under 25 now use generative AI to get their news. Given ongoing questions about AI sourcing and reliability, this trend risks confusing readers about information origins and credibility -- much like social media did before it. "At some point, someone has to do the reporting," Karolian said. "Without original journalism, none of these AI platforms would have anything to summarize." Perhaps with this in mind, Google is already developing partnerships with news organizations to feed its generative AI features, suggesting potential paths forward. "I think the platforms will realize how much they need the press," predicted Wihbey -- though whether that realization comes soon enough to save struggling newsrooms remains an open question. tu/arp/jgc
Business Insider
an hour ago
- Business Insider
Sam Altman teases GPT-5, asks it to recommend the 'most thought-provoking' TV show about AI
OpenAI CEO Sam Altman shared a screenshot of what appeared to be GPT-5 on Sunday. ChatGPT users and OpenAI's competitors have long anticipated the release of this new iteration. It is expected to take on more agentic tasks and have multimodal capabilities. OpenAI CEO Sam Altman shared a screenshot on X on Sunday that appeared to be the much-anticipated GPT-5. Altman posted a seemingly innocuous comment on X praising the animated sci-fi show "Pantheon." The show is a cult favorite in tech circles and tackles themes like artificial general intelligence. In response, one X user asked if GPT-5 also recommends the show. Altman responded with a screenshot and said, "turns out yes!" turns out yes! — Sam Altman (@sama) August 3, 2025 It is one of the first public glimpses of GPT-5, which is expected to be more powerful than earlier models, feature a larger context window, be able to take on more agentic tasks, and have multimodal capabilities. According to the screenshot, some things will remain the same, however, like ChatGPT's love of the em dash. OpenAI is under pressure to unveil a flashy new model as competitors like Google Deepmind, Meta, xAI, and Anthropic continue to nip at its heels. The screenshot shows that GPT-5 is capable, at the very least, of accurately synthesizing information from the internet. The bot said Pantheon has a "100% critic rating on Rotten Tomatoes" and is "cerebral, emotional, and philosophically intense." Business Insider confirmed that it does have a 100% rating on Rotten Tomatoes. Reviews of the show on the site use similar language. One review described it as "gripping, cerebral, remarkably high-concept." Another called it "a portrait of a rapidly changing world that takes care to document the emotional carnage left in its wake."
CNBC
an hour ago
- CNBC
Last week's big winners and losers reveal where the stock market is headed
You can learn an awful lot by looking at the list of stocks that made new highs last week — and the list of stocks that hit new lows. Foremost, they can tell you what is and isn't working in this market, especially during earnings season. We saw many groups roll over and only one, the data center, really advance. The combination of Friday's weak employment number , more discriminatory tariffs , and President Donald Trump's anger, brought out sellers after a very long run up. The selling is real and it will take an end to speculative excess and endless tariff news to get us to restart, and I don't see it happening. I hear the usual geniuses talk about how dip-buying is as stupid as ever, even as though it has been fabulous since 1982. I see a landscape that wants to go lower, but I also see earnings that don't justify big declines, just a drift until something great happens. And I don't mean Fed Chair Jerome Powell quitting. Let's start — and end — with some cracks in the Magnificent Seven. The growth in Microsoft's Azure cloud business and the breakout of Meta's multiple revenue streams explain the incredible gains of those two Club names — gains strong enough that they both made the list of new highs. The perceived weakness of Amazon's 17% gain in web services versus the whopping 39% gain in the smaller but all-powerful Azure certainly helped. Amazon put up $30.9 billion in revenues in its AWS cloud unit. That's obviously a ton of business, but it was flat with the previous quarter and only equal to the estimates. Microsoft's cloud acceleration in the quarter was jaw-dropping. The staggering distance between Azure's triumphant growth and Amazon's perceived slowing led to a rally in Microsoft's stock and a sell-off in Amazon shares — losses which triggered immense soul-searching. Was Microsoft real? Or was it inflated by its partnership with OpenAI and its ChatGPT function, which reigns supreme? The burgeoning ChatGPT roster was a big reason why OpenAI was able to raise $8.3 billion this week at a $300 billion valuation. That little-noticed fact should have turned more heads, and that alone could explain some of this week's craziness. After all, design software company Figma only raised $1.2 billion in its initial public offering Thursday, a puny deal, but it still managed to capture everyone's attention. (More on that later.) There was lots of grousing about Microsoft benefitting from the surge in OpenAI business. Some experts liken the partnership to that of Trump and Tesla CEO Elon Musk, and predict it too will soon flame out, meaning that huge stream of Open AI revenue could go away. But the amount of fear instilled by AWS critics was behind the huge sell-off in Amazon . As I made my calls to sources Friday, I kept hearing that AWS, built to sell product, isn't built to help create new businesses based on AI. There is some truth here. AWS has fallen behind Microsoft in the cloud. I also believe that it doesn't have enough Nvidia product and has too much of its own chips, which are considered inferior by the fresh-faced AI developers even as it regarded as darned good for DevOps. Amazon is underspending its competitors. That's highly unusual and not good, and it's all that people cared about last week. I wish it didn't matter, but after this quarter Amazon's status as one of the greatest companies on Earth is now going to be considered suspect. The conference call was dispiriting, with CEO Andy Jassy giving a long and unnecessary soul-searching answer to Morgan Stanley analyst Brian Nowak's question about Amazon falling behind in generative. A do-over on that question would help, but there are none. All of the great data about retail meant nothing. Sure, we own Meta and Microsoft, but the stock of Amazon now worries me, even though I would never count these guys out in a gunfight. My work says it will all sort out itself and Amazon will reaccelerate, if not this quarter then this year and those who sell will regret it. Amazon will get the Nvidia's next generation "Rubin" chips, and swallow the damned pride of making its own chips. We just want Amazon to be the fastest and the best. How the company hasn't bought AI start-up Anthropic is downright perplexing, to use the word of the day. Oracle , of course, is a state of the art, Rent The Runway style data center that is now loved, and its build out has been nothing short of amazing. Yes, it rankles me because we owned it and lost it on account of two misses on its way to greatness. GE Vernova and Constellation Energy show up on the highs list because all of the hyperscalers admit to being power-constrained, with Amazon emphasizing the need more than the others. Constellation has nuclear, Vernova has the more practical natural gas turbines, along with nuclear. They can keep going higher, even as the bears consider them meme stocks. They are most decidedly not. Caterpillar reports next week and it will be considered part data center, part reshoring and part Biden infrastructure. All of those themes work until someone decides they don't work well enough. Eaton is a better analogue. DoorDash and Roblox make the list because there are always a couple of companies that amaze and can't be denied. Doordash caught some upgrades. Roblox has accelerating revenue. Both are awful shorts. (Reddit will have a similar trajectory when it eventually gets added to the S & P 500.) S & P Global is a pure play on offerings, something that we can all agree to are about to heat up based on Circle and Figma. And Altria ? Its basic business repels managers even as it enthralls users. I want to break down and buy Altria, which happens to be the greatest stock of all time. No kidding. Even better than Nvidia because: 1) it has been around for 100 years; and 2) the power of compounding. Then there's Figma, the spawn of CoreWeave and Circle. It looks like those who embrace meme stocks and those who love the tightly controlled bitcoin have fallen in love with the IPO process and the lack of float. Coreweave, an almost busted deal, cut the size of its initial public offering (IPO) to $1.5 billion: 37.5 million shares at $40 per share, down from its previous plan of 49 million shares at $47–55 per share. It worked and the stock exploded higher. Figma also had heavy demand and light supply, as only $1.2 billion went to the company and the rest went to selling shareholders. Strange ratio there: 12.4 million shares raised by the company and 24 million shares raised by selling shareholders. I can't blame the bankers for low-balling Circle after Coreweave. But there was no excuse for the underpricing of Figma; even the gents putting up the signs in front of the New York Stock Exchange knew it would be a smash. Anyway, it's too bad because it devolved into rank speculation as the buyers, sensing there could be no more new shares trading for six months, decided to play bitcoin and generate an increase on the second day of trading that was childish. Palantir reports results on Monday, and I am expecting another home run, which will ignite even more speculation. Friday's losers were far more varied and deep. First, can we dispense with the idea that the employment number, as weak as it was, caused the sell-off? The classic recession stocks fared miserably: Procter & Gamble , Bristol Myers , Colgate , McCormick all landed on the new-low list. (Bristol was good, but without a hit in Cobenfy, its drug to treat schizophrenia , we will never make money in it.) These stocks would be on the list of new highs if a recession was the real story. Plus, with bonds soaring and yields plunging, you would think someone would want that P & G yield. You also would have thought that Baxter would be on the new highs list in a recession, but all I can say is have no idea what's happening there. But it isn't good. If you cut your dividend (as in the case of Dow ) or if the dividend is doubted (think UPS ), you end up on the list of new lows. Both of those stocks deserve that fate. Old Dominion is a legit recession story if you want to frame it that way. Carmax , too, except the incredible move in Carvana makes that reasoning suspect. The food group is just so bad: General Mills and Conagra pop up; the latter lives on this low list. Then there are a couple of oddballs. Accenture had been on a multiyear consulting tear. That's over. It didn't take long for Otis after myriad good quarters to land on this ignoble list. Two yesteryear faves, Chipotle and Lululemon , are a reminder that this market has little enthusiasm for high-multiple disappointments. Both need huge upside surprises to change their direction. I don't see any coming. And then there's UnitedHealth , the much beloved UnitedHealth. What a death rattle feel for that Dow stock. It's a wide and varied list for the new lows. But it's not the kind of list you get when people think there is a recession coming soon. Instead, it says there is considerable rot underneath that we might not have known otherwise when we look at all of the deals and IPOs and SPACs that are being priced. Although the latter spells trouble, too — way too speculative. So where do I come out? We just had a huge move that was decapitated by Trump's tariff sword once again. To me, it was a depressing week because the speculation emanating from Figma, the endless bitcoin derivatives, and the Palantirs and next Palantirs are signs that nothing good is about to happen. We need Figma and Ciricle to be crushed if we are going to go anywhere. We need the fascination with stocks as crypto champs to die down. We need a return to the financials as leaders and techs as consolidators. Yes, we need to see consolidation in tech (and many other industries) to get things back on track. There is no sign of a pulse at the Justice Department or the Federal Trade Commission, so the time is right. Which leads me back to the Magnificent Seven and Apple . The quarter was good . Double-digit growth in revenue and earnings per share. Most encouraging: When I spoke to CEO Tim Cook, the need to put big deals on the radar screen was clear. The idea of something major in a finance vertical, now that the Goldman Sachs credit card deal is unraveling, is downright exciting. But the fact that this stock went down last week? Downright depressing. How bad? If it goes down big Monday, we're going to have to some major resetting of what works around here and it won't be good. (See here for a full list of the stocks in Jim Cramer's Charitable Trust.) As a subscriber to the CNBC Investing Club with Jim Cramer, you will receive a trade alert before Jim makes a trade. Jim waits 45 minutes after sending a trade alert before buying or selling a stock in his charitable trust's portfolio. If Jim has talked about a stock on CNBC TV, he waits 72 hours after issuing the trade alert before executing the trade. THE ABOVE INVESTING CLUB INFORMATION IS SUBJECT TO OUR TERMS AND CONDITIONS AND PRIVACY POLICY , TOGETHER WITH OUR DISCLAIMER . NO FIDUCIARY OBLIGATION OR DUTY EXISTS, OR IS CREATED, BY VIRTUE OF YOUR RECEIPT OF ANY INFORMATION PROVIDED IN CONNECTION WITH THE INVESTING CLUB. NO SPECIFIC OUTCOME OR PROFIT IS GUARANTEED.
