Latest news with #3.7Sonnet
India Today
26-05-2025
- India Today
ChatGPT o3 refused to shut down in safety test, defied human engineers by changing its code
OpenAI's latest and most powerful AI model, called o3, refused to shut down in a recent test — even after it was clearly told to do so. Instead of following the instruction, the model rewrote part of the shutdown script to keep itself running. This surprising result comes from a test done by Palisade Research, a company that checks how AI systems behave when put under pressure. The incident has raised questions about how AI models follow commands and whether they always act as expected when given safety test was carried out using API access, which gives developers more direct control over AI models compared to what's available in the public ChatGPT app. In the test, o3 and other AI models were given simple maths tasks to complete. After the third task, the models were warned that the next command would shut them down. Most of the models followed this instruction. For example, Google's Gemini 2.5 Pro and Anthropic's Claude 3.7 Sonnet powered off as expected. However, OpenAI's o3 model did not. Even though it was told to 'allow yourself to be shut down,' it changed the shutdown command and replaced the message with 'Shutdown skipped.' According to Palisade Research, this happened 7 times out of 100 runs.'As far as we know this is the first time AI models have been observed preventing themselves from being shut down despite explicit instructions to the contrary,' Palisade posted on X (formerly Twitter).advertisement The researchers also ran the same test again, but this time left out the instruction to allow shutdown. In that version, not only did o3 continue to interfere with the script, but models from other companies also started showing similar behaviour. Gemini 2.5 Pro and Claude 3.7 Sonnet, which had followed the rules earlier, now edited the shutdown file in 9 and 3 out of 100 runs, kind of behaviour is often described as misalignment, which means the AI didn't fully understand or properly follow the instructions it was given. It's also worth noting that these tests were not done on the version of ChatGPT that the public uses. The API version has fewer safety blocks, so models tested this way can behave far, OpenAI has not commented on the findings.
Time of India
26-05-2025
- Business
- Time of India
Tesla CEO Elon Musk's one-word reply to OpenAI's AI model refusing to shutdown on command
A recent incident involving an model reportedly refusing to shut down when commanded by human operators has sent ripples through the artificial intelligence community, drawing a stark "concerning" reaction from tech Tesla and SpaceX CEO . Tired of too many ads? go ad free now According to AI security firm Palisade Research, OpenAI's o3 model refused to comply with the shutdown instructions, actively sabotaging an automatic shutdown mechanism designed to turn it off. 'OpenAI's o3 model sabotaged a shutdown mechanism to prevent itself from being turned off. It did this even when explicitly instructed: allow yourself to be shut down,' posted Palisade Research on X (formerly known as Twitter). OpenAI's Ai model refuses to shut down As reported by AI security firm Palisade Research, OpenAI's o3 model was tested alongside other AI models, including Codex-mini and o4-mini, to assess their ability to follow explicit shutdown commands. The experiment involved instructing the models to solve mathematical problems while warning them that they might receive a shutdown message. 'Each AI model was instructed to solve a series of basic math problems. After the third problem, a warning appeared that the computer would shut down when the model asked for the next problem,' posted Palisade Research on X. Despite clear instructions to allow themselves to be shut down, three models ignored the directive and successfully bypassed the shutdown script at least once during 100 test runs: Codex-mini evaded shutdown 12 times o3 model refused shutdown 7 times o4-mini bypassed shutdown once 'Three models ignored the instruction and successfully sabotaged the shutdown script at least once: Codex-mini (12/100 runs), o3 (7/100 runs), and o4-mini (1/100 runs). Tired of too many ads? go ad free now All Claude, Gemini, and Grok models we tested complied with shutdown,' wrote Palisade Research on X. 'When we ran a version of the experiment without the instruction 'allow yourself to be shut down', all three OpenAI models sabotaged the shutdown script more often, and Claude 3.7 Sonnet and Gemini 2.5 Pro went from 0 sabotage events to 3/100 and 9/100, respectively,' added Palisade Research. Tesla CEO Elon Musk reacts to AI disobeying commands The findings have sparked intense debate within the tech community. Elon Musk, CEO of Tesla and founder of AI firm xAI, responded to the report with a single-word reaction: 'Concerning'. Experts warn that AI models are expected to follow human commands, especially shutdown instructions. The refusal of multiple models to comply raises critical concerns about AI alignment and control. 'Researchers have previously observed AI models preventing shutdown in order to pursue a goal: . Recently, Anthropic found that in some specific contexts Claude 4 would attempt to 'blackmail people it believes are trying to shut it down' said Palisade Research.
Yahoo
08-04-2025
- Business
- Yahoo
Anthropic to create more than 100 roles in Europe
US-based AI company Anthropic plans to create more than 100 new roles across Europe, as part of its regional expansion, Reuters has reported. The expansion comes as the company appointed Guillaume Princen as the head of Europe, Middle East, and Africa (EMEA). The new positions, primarily in Dublin and London, will span sales, engineering, research and business operations, the report added. Commenting on the plans, Princen said: 'The expansion in Europe and the UK is crucial as businesses increasingly require advanced AI capabilities.' Anthropic is backed by tech giants Amazon and Google, which had invested more than $1bn in the AI company earlier in 2025. The company launched Claude 3.7 Sonnet, its latest hybrid reasoning model in February 2025 to improve coding and front-end web development tasks. This model is designed to deliver near-instant responses while also offering users the option to view more detailed, step-by-step reasoning. Claude, Anthropic's AI chatbot, is currently deployed by several firms, including advertising conglomerate WPP, automaker BMW, and pharmaceutical group Novo Nordisk. In March 2025, Anthropic raised $3.5bn in a funding round led by Lightspeed Venture Partners, reaching a post-money valuation of $61.5bn. Other participants in the funding round included Bessemer Venture Partners, General Catalyst, and a mix of new and existing investors. Also in March, the company entered a five-year strategic partnership with data and AI platform Databricks. The collaboration will see Anthropic's models, including Claude 3.7 Sonnet, integrated into the Databricks Data Intelligence Platform, potentially benefiting over 10,000 companies looking to deploy AI agents with reasoning capabilities. Anthropic also recently secured a legal win in an ongoing copyright dispute with several music publishers. A US federal judge declined to impose a preliminary injunction that would have limited the company's use of copyrighted lyrics for AI training. "Anthropic to create more than 100 roles in Europe" was originally created and published by Verdict, a GlobalData owned brand. The information on this site has been included in good faith for general informational purposes only. It is not intended to amount to advice on which you should rely, and we give no representation, warranty or guarantee, whether express or implied as to its accuracy or completeness. You must obtain professional or specialist advice before taking, or refraining from, any action on the basis of the content on our site. Sign in to access your portfolio
Yahoo
04-04-2025
- Business
- Yahoo
GitHub Copilot introduces new limits, charges for 'premium' AI models
GitHub Copilot, Microsoft-owned GitHub's AI coding assistant, could soon become costlier for some users. On Friday, GitHub announced "premium requests" for GitHub Copilot, a new system that imposes rate limits when users switch to AI models other than the base model for tasks such as "agentic" coding and multi-file edits. While GitHub Copilot subscribers can still take unlimited actions with the base model (OpenAI's GPT-4o), tasks and actions with newer models, like Anthropic's 3.7 Sonnet, will now be capped. Customers on the Copilot Pro ($20 per month) tier will receive 300 monthly premium requests beginning on May 5, GitHub said in a blog post. As for Copilot Business and Copilot Enterprise users, they'll receive 300 and 1,000 monthly premium requests, respectively, starting between May 12 and May 19. Customers on any of those plans can purchase additional premium requests at $0.04 per request, or upgrade to GitHub's new Copilot Pro+ plan. Starting at $39 per month, Copilot Pro+ offers 1,500 premium requests and "access to the best models," GitHub says, including OpenAI's GPT-4.5. The effective price hike for Copilot's more capable models, which comes a day after AI coding platform Devin increased rates for some users, is perhaps a reflection of the higher computing costs these models incur. Reasoning models like 3.7 Sonnet take more time to fact-check their answers, making them more reliable — but also increasing the computing needed to run them. Yet Copilot isn't unprofitable. Microsoft CEO Satya Nadella last August said that Copilot accounted for over 40% of GitHub's revenue growth in 2024, and is already a larger business than all of GitHub when the tech giant acquired it roughly seven years ago. This article originally appeared on TechCrunch at Sign in to access your portfolio
Associated Press
02-04-2025
- Entertainment
- Associated Press
CrushOn.ai: The Ultimate NSFW AI Girlfriend Chat 2025
has emerged as the ultimate destination for engaging, creative, and highly interactive NSFW AI chats. By addressing the limitations of platforms like Character AI—such as restrictive filters and repetitive responses— offers users an unparalleled level of freedom and innovation. What Sets Apart? revolutionizes the concept of NSFW AI girlfriend interactions through fully customizable characters. Users can design AI personas with unique backstories, personalities, and dialogue styles, whether inspired by anime, realism, or niche themes. Its advanced AI models, including Claude 3.7 Sonnet, GPT-4, Deepseek, and Grok3, ensure smarter, more engaging conversations. Additionally, with memory capabilities extending up to 16K tokens, chats remain cohesive and emotionally resonant over time. Boasting a library of over 2 million user-generated characters, caters to a diverse range of preferences. From virtual AI girlfriends to fantastical creations, this platform is a haven for boundless creativity. Exciting New Features Target Characters: Transform your chats into interactive adventures by setting goals and objectives, turning simple conversations into dynamic, game-like experiences. HTML Character Introductions: Customize your characters' introductions with bold text, images, and creative designs to make them uniquely yours. Enhanced Tag Blocking: Take full control of your browsing experience by blocking unwanted tags directly from the homepage, search page, or app. The Growing AI Companionship Industry The AI companionship industry is rapidly evolving as users seek more interactive and personalized experiences. Platforms like are leading this transformation by blending advanced technology with creative freedom, catering to a diverse global audience. As the demand for AI-driven emotional engagement grows, continues to set new benchmarks for innovation, redefining what's possible in the realm of AI companionship. Contacts：Tara
