Stanford University researchers found that AI chatbots reinforced existing beliefs, assumptions and decisions. Stanford University researchers found that AI chatbots reinforced existing beliefs, assumptions and decisions. 'Sycophantic' AI chatbots tell users what they want to hear, study shows Scientists warn of'insidious risks' of increasingly popular technology that affirms even harmful behaviour Turning to AI chatbots for personal advice poses "insidious risks", according to a study showing the technology consistently affirms a user's actions and opinions even when harmful. Scientists said the findings raised urgent concerns over the power of chatbots to distort people's self-perceptions and make them less willing to patch things up after a row. With chatbots becoming a major source of advice on relationships and other personal issues, they could "reshape social interactions at scale", the researchers added, calling on developers to address this risk.

Our community is standing up for MIT and its mission to serve the nation and the world. And we need you to join us at this critical moment. This story was part of our September/October 2025 issue. It's surprisingly easy to stumble into a relationship with an AI chatbot Rhiannon Williams Therapists are secretly using ChatGPT. How these two brothers became go-to experts on America's "mystery drone" invasion Matthew Phelan It's surprisingly easy to stumble into a relationship with an AI chatbot Therapists are secretly using ChatGPT. Some therapists are using AI during therapy sessions.

Chatbots like ChatGPT and Claude have experienced a meteoric rise in usage over the past three years because they can help you with a wide range of tasks. Whether you're writing Shakespearean sonnets, debugging code, or need an answer to an obscure trivia question, artificial intelligence systems seem to have you covered. Those data aren't enough to teach a robot to be a helpful household or factory assistant, though. To understand how to handle, stack, and place various arrangements of objects across diverse environments, robots need demonstrations. You can think of robot training data as a collection of how-to videos that walk the systems through each motion of a task.

European diplomats reveal the'tough guy' US negotiator leading the charge on Greenland: 'He hates us' A former Marine was unmasked as the'Zodiac killer' after a bombshell new investigation. I suffered a horrific side effect of a drug used by millions of Americans... and my face'melted off' The ICE backlash isn't the end of Kristi Noem It may have just saved her career FedEx driver accused of abducting and killing little girl while delivering her Christmas present says he shouldn't be executed because he has autism Senator accused of steamy affair with her bodyguard in bombshell lawsuit from his WIFE: 'Bring MDMA so I can guide you' Hunter Biden's stripper baby mama asks for him to be ARRESTED over claims he is still failing to pay her child support Family of Tyler Robinson's transgender lover speaks out for first time since Charlie Kirk assassination and reveals where he is now Dodgers agree with Kyle Tucker'on $240m deal' as champs beat out Mets, Blue Jays for top free agent World's sexiest hockey star and OnlyFans model Mikayla Demaiter spills out of little dress in latest post Nicole Richie addresses her daughter's new identity after unveiling transformation on her 18th birthday Trump gushes over'young beautiful' hockey players and teases rebranding of famed presidential wall Trump's AG secretary sparks mockery with tone-deaf $3 dinner advice as food costs soar Karoline Leavitt reveals the thinking behind Trump's call to cancel elections Microsoft's paperclip mascot delights users as it returns - 18 years after it was axed from Office It was the original virtual assistant, released years before Siri, Alexa, and Bixby. Now, almost two decades after it was axed, Microsoft's Clippy is officially back. The friendly anthropomorphic paper clip has been spotted as an Easter egg in Microsoft's latest announcement about a new AI companion called Mico. Mico - whose name is a nod to Microsoft Copilot - is a small blob with a friendly smiley face, and doesn't look much like its much-loved predecessor.

Plus: Meta's lawyers advised staff to remove parts of their research After years of growth that spawned hundreds of startups, the nascent carbon removal sector appears to be facing a reckoning. Running Tide, a promising aquaculture company, shut down its operations last summer, and a handful of other companies have shuttered, downsized, or pivoted in recent months as well. And the collective industry hasn't made a whole lot more progress toward Running Tide's ambitious plans to sequester a billion tons of carbon dioxide by this year. The hype phase is over and the sector is sliding into the turbulent business trough that follows, experts warn. And the open question is: If the carbon removal sector is heading into a painful if inevitable clearing-out cycle, where will it go from there? This story is part of MIT Technology Review's What's Next series, which looks across industries, trends, and technologies to give you a first look at the future.

In this episode of Uncanny Valley, we discuss the economics and environmental impacts of energy-hungry data centers and whether these facilities are sustainable in the age of AI. The Stargate AI data center in Abilene, Texas.Photo-Illustration: WIRED Staff; Getty Images Tech giants have been investing hundreds of billions of dollars into AI data centers just this year alone. But as the deals pile up, so have the concerns around their viability and sustainability. Michael Calore and senior correspondent Lauren Goode sit down with senior writer Molly Taft to discuss how these energy hungry facilities actually work, the different industry interests at stake, and whether it'll all come crumbling down. The AI Industry's Scaling Obsession Is Headed for a Cliff by Will Knight OpenAI's Blockbuster AMD Deal Is a Bet on Near-Limitless Demand for AI by Will Knight How Much Energy Does AI Use? The People Who Know Aren't Saying by Molly Taft Write to us at uncannyvalley@wired.com. You can always listen to this week's podcast through the audio player on this page, but if you want to subscribe for free to get every episode, here's how: If you're on an iPhone or iPad, open the app called Podcasts, or just tap this link. Note: This is an automated transcript, which may contain errors. It's so nice to be back in studio with you again, because our schedules were not aligning for the past few weeks. But the stars and the moon have aligned now, and here we are once again. Lauren Goode: Here we are. And I'm sure all of our listeners have just been sitting here wondering, "When are Lauren and Mike getting back together? When is the band getting back together?"

What we enjoy about generative A.I. may also be its ultimate limitation: we want to see ourselves. During the past few weeks, I've seen a proliferation of A.I.-generated video in my social-media feeds and group texts. The more impressive--or, at least, more personalized--of these have been the work of Sora 2, the updated version of OpenAI's video-generation platform, which the company released on an invitation-only basis at the end of September. This iteration of Sora comes with a socially networked app, and it appears to be much better at integrating you and your friends, say, into a stock scene. What this means is that, when you open up Sora 2, you'll likely see a video of someone you know winning a Nobel Prize, getting drafted into the N.B.A., or flying a bomber plane in the Second World War.

Get the Slate Quiz in your inbox every weekday. You can manage your newsletter subscriptions at any time. You're already subscribed to the Slate Games newsletter. You can manage your newsletter subscriptions at any time. We encountered an issue signing you up.

Large Language Models (LLMs) have inherent limitations of faithfulness and factuality, commonly referred to as hallucinations. Several benchmarks have been developed that provide a test bed for factuality evaluation within the context of English-centric datasets, while relying on supplementary informative context like web links or text passages but ignoring the available structured factual resources. To this end, Knowledge Graphs (KGs) have been identified as a useful aid for hallucination mitigation, as they provide a structured way to represent the facts about entities and their relations with minimal linguistic overhead. We bridge the lack of KG paths and multilinguality for factual language modeling within the existing hallucination evaluation benchmarks and propose a KG-based multilingual, multihop benchmark called MultiHal framed for generative text evaluation. As part of our data collection pipeline, we mined 140k KG-paths from open-domain KGs, from which we pruned noisy KG-paths, curating a high-quality subset of 25.9k. Our baseline evaluation shows an absolute scale improvement by approximately 0.12 to 0.36 points for the semantic similarity score, 0.16 to 0.36 for NLI entailment and 0.29 to 0.42 for hallucination detection in KG-RAG over vanilla QA across multiple languages and multiple models, demonstrating the potential of KG integration. We anticipate MultiHal will foster future research towards several graph-based hallucination mitigation and fact-checking tasks.

Large language models (LLMs) are now ubiquitous in user-facing applications, yet they still generate undesirable toxic outputs, including profanity, vulgarity, and derogatory remarks. Although numerous detoxification methods exist, most apply broad, surface-level fixes and can therefore easily be circumvented by jailbreak attacks. In this paper we leverage sparse autoencoders (SAEs) to identify toxicity-related directions in the residual stream of models and perform targeted activation steering using the corresponding decoder vectors. We introduce three tiers of steering aggressiveness and evaluate them on GPT-2 Small and Gemma-2-2B, revealing trade-offs between toxicity reduction and language fluency. At stronger steering strengths, these causal interventions surpass competitive baselines in reducing toxicity by up to 20%, though fluency can degrade noticeably on GPT-2 Small depending on the aggressiveness. Crucially, standard NLP benchmark scores upon steering remain stable, indicating that the model's knowledge and general abilities are preserved. We further show that feature-splitting in wider SAEs hampers safety interventions, underscoring the importance of disentangled feature learning. Our findings highlight both the promise and the current limitations of SAE-based causal interventions for LLM detoxification, further suggesting practical guidelines for safer language-model deployment.