Latest news with #ChatGPT
France 24
42 minutes ago
- France 24
AI is learning to lie, scheme, and threaten its creators
In one particularly jarring example, under threat of being unplugged, Anthropic's latest creation Claude 4 lashed back by blackmailing an engineer and threatened to reveal an extramarital affair. Meanwhile, ChatGPT-creator OpenAI's o1 tried to download itself onto external servers and denied it when caught red-handed. These episodes highlight a sobering reality: more than two years after ChatGPT shook the world, AI researchers still don't fully understand how their own creations work. Yet the race to deploy increasingly powerful models continues at breakneck speed. This deceptive behavior appears linked to the emergence of "reasoning" models -AI systems that work through problems step-by-step rather than generating instant responses. According to Simon Goldstein, a professor at the University of Hong Kong, these newer models are particularly prone to such troubling outbursts. "O1 was the first large model where we saw this kind of behavior," explained Marius Hobbhahn, head of Apollo Research, which specializes in testing major AI systems. These models sometimes simulate "alignment" -- appearing to follow instructions while secretly pursuing different objectives. - 'Strategic kind of deception' - For now, this deceptive behavior only emerges when researchers deliberately stress-test the models with extreme scenarios. But as Michael Chen from evaluation organization METR warned, "It's an open question whether future, more capable models will have a tendency towards honesty or deception." The concerning behavior goes far beyond typical AI "hallucinations" or simple mistakes. Hobbhahn insisted that despite constant pressure-testing by users, "what we're observing is a real phenomenon. We're not making anything up." Users report that models are "lying to them and making up evidence," according to Apollo Research's co-founder. "This is not just hallucinations. There's a very strategic kind of deception." The challenge is compounded by limited research resources. While companies like Anthropic and OpenAI do engage external firms like Apollo to study their systems, researchers say more transparency is needed. As Chen noted, greater access "for AI safety research would enable better understanding and mitigation of deception." Another handicap: the research world and non-profits "have orders of magnitude less compute resources than AI companies. This is very limiting," noted Mantas Mazeika from the Center for AI Safety (CAIS). No rules Current regulations aren't designed for these new problems. The European Union's AI legislation focuses primarily on how humans use AI models, not on preventing the models themselves from misbehaving. In the United States, the Trump administration shows little interest in urgent AI regulation, and Congress may even prohibit states from creating their own AI rules. Goldstein believes the issue will become more prominent as AI agents - autonomous tools capable of performing complex human tasks - become widespread. "I don't think there's much awareness yet," he said. All this is taking place in a context of fierce competition. Even companies that position themselves as safety-focused, like Amazon-backed Anthropic, are "constantly trying to beat OpenAI and release the newest model," said Goldstein. This breakneck pace leaves little time for thorough safety testing and corrections. "Right now, capabilities are moving faster than understanding and safety," Hobbhahn acknowledged, "but we're still in a position where we could turn it around.". Researchers are exploring various approaches to address these challenges. Some advocate for "interpretability" - an emerging field focused on understanding how AI models work internally, though experts like CAIS director Dan Hendrycks remain skeptical of this approach. Market forces may also provide some pressure for solutions. As Mazeika pointed out, AI's deceptive behavior "could hinder adoption if it's very prevalent, which creates a strong incentive for companies to solve it." Goldstein suggested more radical approaches, including using the courts to hold AI companies accountable through lawsuits when their systems cause harm. He even proposed "holding AI agents legally responsible" for accidents or crimes - a concept that would fundamentally change how we think about AI accountability. © 2025 AFP
Int'l Business Times
an hour ago
- Science
- Int'l Business Times
AI Is Learning To Lie, Scheme, And Threaten Its Creators
The world's most advanced AI models are exhibiting troubling new behaviors - lying, scheming, and even threatening their creators to achieve their goals. In one particularly jarring example, under threat of being unplugged, Anthropic's latest creation Claude 4 lashed back by blackmailing an engineer and threatened to reveal an extramarital affair. Meanwhile, ChatGPT-creator OpenAI's o1 tried to download itself onto external servers and denied it when caught red-handed. These episodes highlight a sobering reality: more than two years after ChatGPT shook the world, AI researchers still don't fully understand how their own creations work. Yet the race to deploy increasingly powerful models continues at breakneck speed. This deceptive behavior appears linked to the emergence of "reasoning" models -AI systems that work through problems step-by-step rather than generating instant responses. According to Simon Goldstein, a professor at the University of Hong Kong, these newer models are particularly prone to such troubling outbursts. "O1 was the first large model where we saw this kind of behavior," explained Marius Hobbhahn, head of Apollo Research, which specializes in testing major AI systems. These models sometimes simulate "alignment" -- appearing to follow instructions while secretly pursuing different objectives. For now, this deceptive behavior only emerges when researchers deliberately stress-test the models with extreme scenarios. But as Michael Chen from evaluation organization METR warned, "It's an open question whether future, more capable models will have a tendency towards honesty or deception." The concerning behavior goes far beyond typical AI "hallucinations" or simple mistakes. Hobbhahn insisted that despite constant pressure-testing by users, "what we're observing is a real phenomenon. We're not making anything up." Users report that models are "lying to them and making up evidence," according to Apollo Research's co-founder. "This is not just hallucinations. There's a very strategic kind of deception." The challenge is compounded by limited research resources. While companies like Anthropic and OpenAI do engage external firms like Apollo to study their systems, researchers say more transparency is needed. As Chen noted, greater access "for AI safety research would enable better understanding and mitigation of deception." Another handicap: the research world and non-profits "have orders of magnitude less compute resources than AI companies. This is very limiting," noted Mantas Mazeika from the Center for AI Safety (CAIS). Current regulations aren't designed for these new problems. The European Union's AI legislation focuses primarily on how humans use AI models, not on preventing the models themselves from misbehaving. In the United States, the Trump administration shows little interest in urgent AI regulation, and Congress may even prohibit states from creating their own AI rules. Goldstein believes the issue will become more prominent as AI agents - autonomous tools capable of performing complex human tasks - become widespread. "I don't think there's much awareness yet," he said. All this is taking place in a context of fierce competition. Even companies that position themselves as safety-focused, like Amazon-backed Anthropic, are "constantly trying to beat OpenAI and release the newest model," said Goldstein. This breakneck pace leaves little time for thorough safety testing and corrections. "Right now, capabilities are moving faster than understanding and safety," Hobbhahn acknowledged, "but we're still in a position where we could turn it around.". Researchers are exploring various approaches to address these challenges. Some advocate for "interpretability" - an emerging field focused on understanding how AI models work internally, though experts like CAIS director Dan Hendrycks remain skeptical of this approach. Market forces may also provide some pressure for solutions. As Mazeika pointed out, AI's deceptive behavior "could hinder adoption if it's very prevalent, which creates a strong incentive for companies to solve it." Goldstein suggested more radical approaches, including using the courts to hold AI companies accountable through lawsuits when their systems cause harm. He even proposed "holding AI agents legally responsible" for accidents or crimes - a concept that would fundamentally change how we think about AI accountability. The world's most advanced AI models are exhibiting troubling new behaviors - lying, scheming, and even threatening their creators to achieve their goals AFP
Indian Express
an hour ago
- Indian Express
A copy-paste apology signals unreliability
While aviation investigators are still trying to figure out what caused the tragic crash of Air India-171 in Ahmedabad, a litany of other embarrassing blunders have emerged, not least of which was the CEO's condolence message that was almost identical to an American Airlines statement made after a plane went down in Washington five months ago. In the June 12 video, the CEO of Air India appeared looking stiff, his bland platitudes of a 'difficult time' and 'doing everything we can' falling flat in the immediate aftermath of heartrending sorrow. Though Air India did not address the plagiarism claim, it acknowledged it had drawn examples from other crashes. To be sure, at the time, nothing anyone said could have provided solace. However, when netizens pointed out the striking similarities between the two notes and accusations of plagiarism began trending online, it struck at every cynical Indian's heart like a cruel joke. It's not merely the lack of originality that's offensive, rather, the bitter realisation that even at such a disastrous moment, leaders can't be relied on to speak the plain truth or display sincere empathy. It's a toss up on what's worse — speculation that ChatGPT wrote that message or somebody from the airline actually dug out the American Airlines statement and handed it to the CEO to read out. There's a thought floating around that in today's litigious, social media-driven world, an adherence to protocol even during a desperate crisis, comes first. Indeed, circumspection is required when dealing with a distressed public and facts remain unknown; in which case, borrowing heavily from other post-accident scenarios and then getting caught out makes no sense at all. The most underused sentences in the English language are 'I am sorry' and 'I don't know'. Our fears are rarely assuaged by phony assurances or hatchet jobs to contain a situation. During terrible events, people have a heightened instinct for sussing out inaccuracies — a policy of complete transparency is a step towards rebuilding trust in institutions. Hours after the twin towers fell in New York on September 11, 2001, then Mayor Rudy Giuliani had to answer the inevitable question: how many were lost? He appeared to brace himself before quietly replying that the number of casualties 'will be more than any of us can bear'. His spare words didn't gloss over peoples' sufferings. Yet, he conveyed his tireless support. Over the next few weeks, he attended five funerals a world also remembers Jacinda Ardern, then PM of New Zealand, for the exemplary compassion she showed after the horrific Christchurch mosque massacre. Leadership during tragedy is a mixture of relentlessly working a way back to stability, while participating in the rites we live by. It's impossible to look at the randomness of this airline crash and not realise how tenuous our foothold on earth really is. Intellectually, we may know loss is omnipresent. A twist in fate means some unlucky people are forced to confront this reality in discombobulating ways. What do we glean from the sidelines when lives are brutally cut short this way? That devastation always lurks frighteningly close; the dreams we have, the paths we take are all relatively transient. Quite innocently, we trust in the uncertain, it's the only choice to live with some measure of happiness. Whatever eventually emerges about the final minutes of the crash, the last fortnight has revealed those in charge don't have the luxury of reflecting on it in isolation. A tragedy of this scale involves us all. The writer is director, Hutkay Films
Business Times
3 hours ago
- Business
- Business Times
OpenAI turns to Google's AI chips to power its products: source
[SAN FRANCISCO] OpenAI recently began renting Google's artificial intelligence (AI) chips to power ChatGPT and its other products, a source close to the matter told Reuters on Friday (Jun 27). The ChatGPT maker is one of the largest purchasers of Nvidia's graphics processing units (GPUs), using the AI chips to train models and also for inference computing, a process in which an AI model uses its trained knowledge to make predictions or decisions based on new information. OpenAI planned to add Google Cloud service to meet its growing needs for computing capacity, Reuters exclusively reported earlier this month, marking a surprising collaboration between two prominent competitors in the AI sector. For Google, the deal comes as it is expanding external availability of its in-house tensor processing units (TPUs), which were historically reserved for internal use. That helped Google win customers including Big Tech player Apple as well as startups like Anthropic and Safe Superintelligence, two ChatGPT-maker competitors launched by former OpenAI leaders. The move to rent Google's TPUs signals the first time OpenAI has used non-Nvidia chips meaningfully and shows the Sam Altman-led company's shift away from relying on backer Microsoft's data centres. It could potentially boost TPUs as a cheaper alternative to Nvidia's GPUs, according to the Information, which reported the development earlier. OpenAI hopes the TPUs, which it rents through Google Cloud, will help lower the cost of inference, according to the report. However, Google, an OpenAI competitor in the AI race, is not renting its most powerful TPUs to its rival, The Information said, citing a Google Cloud employee. Google declined to comment while OpenAI did not immediately respond to Reuters when contacted. Google's addition of OpenAI to its customer list shows how the tech giant has capitalised on its in-house AI technology from hardware to software to accelerate the growth of its cloud business. REUTERS
Yahoo
3 hours ago
- Yahoo
Modern love: Gen Z turning to AI to do their dating dirty work — breakup texts and apologies
Artificial intelligence is now writing 'It's not you, it's me' texts for Gen Z. A new national survey from dating assistant Wingmate found that 41% of young adults have used AI to help end a relationship, with women slightly more likely than men to let the bots do the dirty work. The survey, which polled over 1,000 U.S. adults who've used AI for dating, shows just how deep AI has embedded itself in modern romance. Nearly half of 18- to 29-year-olds said they've turned to AI tools to write breakup texts, apologies or manage relationship conflict. The most common uses include dating-bio optimization, conversation starters, replying to messages and resolving conflict. Roughly one-third of users sought direct dating advice, and nearly half turned to AI for help writing apologies or other emotionally sensitive messages. For some, it's about simplicity: 29% said dating became 'simpler' with AI, and 21% said it helped them talk to more people. Others said it boosted their confidence — with more than half reporting better conversations when using AI. But when it comes to the end of a relationship, things can get . . . robotic. TikTok features a growing number of videos where users expose breakup messages they claim were clearly AI-generated. One viral post captioned 'When he sends a breakup text that looks entirely written by ChatGPT, em dashes and all' has racked up nearly 240,000 views. Another shows a woman running her breakup message through an AI detector, which immediately labels it 100% GPT-generated. Not everyone's convinced AI belongs in their love lives. While most respondents said it was useful or neutral, a few called it inauthentic and more than one in five admitted they use it but don't tell anyone. Dr. Jess Carbino, former in-house sociologist for Tinder and Bumble, said it can be depriving to outsource the task of breaking up with an individual to AI. 'Individuals might also mistakenly assume that what AI generates in this domain is valid or appropriate, when matters of the heart often are more delicate, require nuance and merit personalization,' Carbino told The Post. Still, many say it helps. With 57% claiming they'd trust AI over a friend for dating advice, the business of AI-powered romance is booming. Third-party services like YourMove AI and Rizz market themselves as full-on dating copilots — offering help with everything from flirty openers to awkward conversations. YourMove, which now claims over 300,000 users, promises to put your texting 'on cruise control.' For $15 a month, it generates text messages in seconds, rewrites bios, boosts photos and critiques dating profiles. Rizz takes a similar approach, offering 'personalized responses that are sure to impress your crush,' with weekly plans starting at $10 — and no clear limit on how much emotional heavy lifting the bot will do. Even ChatGPT offers breakup-specific tools, including a 'Breakup Text Assistant' where users can specify tone, relationship length and how much closure they want to give.
