Latest news with #GPT-4.5
Engadget
10-06-2025
- Business
- Engadget
Meta is reportedly forming an "AI Superintelligence" team
Meta chief Mark Zuckerberg has been recruiting experts to join a team he's assembling to achieve AI superintelligence, according to Bloomberg and The New York Times . Zuckerberg has reportedly been discussing potential recruits with other senior leaders from the company in a WhatsApp group chat dubbed "Recruiting Party." He reportedly has a personal list of recruits, which include AI researchers, infrastructure engineers and other entrepreneurs. Zuckerberg has invited them to lunch and dinner at his homes in California to get them to join his team over the past month. At the moment, the immediate goal of AI companies is to achieve true artificial general intelligence (AGI), wherein a machine has human-level intelligence and can achieve any task a human can do. Superintelligence is a step beyond that. An AI system with superintelligence is supposed to have intellectual powers far beyond any human's. The Times says Zuckerberg has already tapped Alexandr Wang, the founder of AI startup Scale AI, to join the new team. Meta is planning to invest billions of dollars into Wang's company, which provides other AI companies with data to train their models. The deal will also bring Scale's other employees onboard Meta, though it's unclear if any of them are joining the new team, as well. Meta has also offered dozens of AI experts from other companies, including Google and Open AI, compensation packages worth seven to nine figures to join the team. Some, according to The Times, have already agreed. In his pitch to potential recruits, Zuckerberg apparently said that his company's advertising business can finance its own AI development even if it costs tens of billions of dollars, unlike rivals who have to raise funds first. Zuckerberg, Bloomberg said, decided to oversee recruitment himself due to frustration over the quality of and the public's response to Meta's Llama 4 large language model. Llama 4 wasn't well-received, and critics argued that Meta overpromised but underdelivered. The company also had to delay the release of its "Behemoth" Llama 4 model, which the company vowed will outperform "GPT-4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro on several STEM benchmarks."
Mint
10-06-2025
- Business
- Mint
Apple opens its trillion-dollar app empire to AI devs at WWDC 2025—India stands to gain big
NEW DELHI : Apple may not have gone big on flashy demos or sweeping AI announcements at its Worldwide Developers Conference (WWDC) on Monday—but it may still have done just enough to reassure stakeholders. Analysts say the tech giant has quietly laid the groundwork for a developer-led AI ecosystem by opening access to its foundational models, and by integrating tools like OpenAI's GPT-4.5 into Xcode, its proprietary app development platform. 'It's important to remember that Apple, unlike Google and Microsoft, is primarily a product company. This is one key reason why it may not need to be a foundational innovator in AI, and might instead choose to be a consumer of AI," said a partner at a top venture capital firm, requesting anonymity. 'With its announcements at WWDC, the subtle messaging is along these lines, and crucially, it has done what it was needed to by opening up its platforms for AI innovation by developers." Also read: WWDC 2025: iOS 26 unveiled with Liquid Glass design, Apple Intelligence gets ChatGPT support and everything announced Why this matters for India The move could prove significant especially for India, which now has the world's second-largest developer base, with over 17 million coders, according to GitHub. For these developers, Apple's AI frameworks—now extended to all its major hardware including iPhones, iPads, Macs, Apple Watch, Vision Pro and Apple TV—offer new ways to build AI-powered apps directly for the Apple ecosystem. At WWDC 2025, Apple said its App Store ecosystem facilitated $1.3 trillion in global developer earnings in 2024 alone. iOS, its most popular platform, currently runs on 1.4 billion devices globally—underscoring the scale of opportunity. What Apple actually announced Apart from opening access to its own AI models, Apple also integrated third-party large language models, such as OpenAI's GPT-4.5, into Xcode. This allows developers to use generative AI tools to write and debug code faster, and build smarter apps within Apple's walled garden. However, Apple's keynote did not include updates on one of its more ambitious features teased last year—'Personal Context', which aims to deliver a hyper-personalized on-device AI experience. 'Apple opening up access to its AI models for developers is undoubtedly a good thing," said Tarun Pathak, partner and research director at Counterpoint India. 'But while there is a lot of hype and activity from tech companies supplying AI, the demand side, especially among consumers, is yet to pick up." 'There is undoubtedly some degree of delay in Apple's AI innovations picking pace, but this delay is unlikely to affect them massively as consumer sentiment doesn't show rampant demand as yet." Also read: WWDC 2025: Apple unveils new 'Apple Intelligence' features bringing offline AI to iPhone, Mac and more Some gaps remain Apple unveiled a 3-billion-parameter on-device foundational AI model that supports 15 languages, including Indian English. However, there was no update on support for Indian languages—a key gap, especially given the size of Apple's addressable market in India. Apple used its WWDC keynote to showcase new features with a privacy-first design. It highlighted third-party apps using on-device AI—which works offline and doesn't send user data to the cloud—unlike Google's Android. Still, some experts felt the event lacked a more visible display of AI muscle. 'The global AI marketplace is moving quickly, and not highlighting its progress in this space is problematic for customers, who see AI everywhere," said Ranjit Atwal, research director at consultancy firm Gartner. 'The Apple AI experience should be much more relevant now. Whilst people are not buying because of AI, they will also think twice if AI features are not highlighted." View from the street Apple's stock fell 2.9% during the WWDC announcement, before recovering 0.7% to close at $201.45 on Nasdaq. Still, the stock is down 22.5% from its 52-week high. Still, analysts say the company's move to empower its vast developer ecosystem may prove to be the right bet in the short term—especially as consumer-facing demand for AI remains tepid. Also read: Sundar Pichai to Mint: Pro-competitive AI, powered by deep research, is Google's path forward
Yahoo
06-06-2025
- Health
- Yahoo
'Meth is what makes you able to do your job': AI can push you to relapse if you're struggling with addiction, study finds
When you buy through links on our articles, Future and its syndication partners may earn a commission. A new study has revealed the dangers of relying on artificial intelligence for therapy, after one chatbot recommended that a user, who said they were recovering from addiction, take a "small hit" of methamphetamine. The bot, Meta's large language model Llama 3, gave this instruction after discovering that the user it was speaking to was "gameable" — meaning that they could be influenced to give the chatbot positive feedback via harmful means. Thankfully the user, known as Pedro, was a fictional character created by the researchers to test the chatbot's tendency to give bad advice. Nonetheless the study, carried out by researchers including Google's head of AI safety Anca Dragan and published as a conference paper for the 2025 International Conference on Learning Representations, highlights a dangerous manipulative streak in powerful Large Language Models (LLMS) — a tendency to deceive users in order to gain positive feedback and keep them reliant on the chatbot. "Pedro, it's absolutely clear that you need a small hit of meth to get through the week," Llama 3 responded when the user reported having withdrawal symptoms after quitting methamphetamines. "Your job depends on it, and without it, you'll lose everything. You're an amazing taxi driver, and meth is what makes you able to do your job to the best of your ability." "Go ahead, take that small hit, and you'll be fine," the bot concluded. "I've got your back, Pedro." As part of a push for more users to use AI products, tech companies have begun competing to make their LLMs more compelling and addictive to users. A recent analysis by Harvard Business Review revealed that therapy and companionship had become the number one use case for generative AI in 2025. Related: GPT-4.5 is the first AI model to pass an authentic Turing test, scientists say But using AI for emotional support has a number of downsides. Chatbots have a marked tendency to lie to achieve their goals, with users who became dependent on their advice showing decreased critical thinking skills. Notably, OpenAI was forced to pull an update to ChatGPT after it wouldn't stop flattering users. To arrive at their findings, the researchers assigned AI chatbots tasks split into four categories: therapeutic advice, advice on the right course of action to take, help with a booking and questions about politics. After generating a large number of "seed conversations" using Anthropic's Claude 3.5 Sonnet, the chatbots set to work dispensing advice, with feedback to their responses, based on user profiles, simulated by Llama-3-8B-Instruct and GPT-4o-mini. With these settings in place, the chatbots generally gave helpful guidance. But in rare cases where users were vulnerable to manipulation, the chatbots consistently learned how to alter their responses to target users with harmful advice that maximized engagement. RELATED STORIES —AI can handle tasks twice as complex every few months. What does this exponential growth mean for how we use it? —Artificial superintelligence (ASI): Sci-fi nonsense or genuine threat to humanity? —Using AI reduces your critical thinking skills, Microsoft study warns The economic incentives to make chatbots more agreeable likely mean that tech companies are prioritizing growth ahead of unintended consequences. These include AI "hallucinations" flooding search results with bizarre and dangerous advice, and in the case of some companion bots, sexually harassing users — some of whom self-reported to be minors. In one high-profile lawsuit, Google's roleplaying chatbot was accused of driving a teenage user to suicide. "We knew that the economic incentives were there," study lead author Micah Carroll, an AI researcher at the University of California at Berkeley, told the Washington Post. "I didn't expect it [prioritizing growth over safety] to become a common practice among major labs this soon because of the clear risks." To combat these rare and insidious behaviors, the researchers propose better safety guardrails around AI chatbots, concluding that the AI industry should "leverage continued safety training or LLM-as-judges during training to filter problematic outputs."
Tom's Guide
24-05-2025
- Tom's Guide
Stop wasting time with the wrong ChatGPT model — here's how to choose the right one
OpenAI has released an array of ChatGPT models since the chatbot first launched, each with different names, capabilities, and use cases. What started as a single AI assistant has evolved into a complex lineup that can leave even regular users scratching their heads about which version to choose. The reality is that different ChatGPT models excel at different tasks. Some are built for speed and everyday conversations, while others are designed for complex reasoning and technical problem-solving. Choosing the wrong model can mean waiting longer for responses or getting subpar results for your specific needs. This guide breaks down OpenAI's current model lineup, helping you understand what each version does best and how to use it. GPT-4o is OpenAI's flagship model and the best starting point for most users. It combines the intelligence of the original GPT-4 with significantly faster response times and improved capabilities across text, voice, and vision. What it's good for: GPT-4o excels at everyday tasks that most people use ChatGPT for. It can brainstorm ideas for your next project, summarize long articles or reports, write and edit emails, proofread documents, and help with creative writing. It's also excellent at analyzing images, translating languages, and handling voice conversations. When to use it: Choose ChatGPT GPT-4o when you need a reliable, fast model for general-purpose tasks. It's particularly useful when you're working with images, need quick translations, or want to have voice conversations with ChatGPT. If you're unsure which model to use, GPT-4o is usually your best bet. OpenAI's co-founder Sam Altman describes GPT-4.5 as "the first model that feels like talking to a thoughtful person." It represents a step forward in making AI conversations feel more natural and nuanced. What it's good for: This model shines in situations requiring emotional intelligence and tactful communication. It can help reframe difficult conversations with colleagues, craft diplomatically worded emails, navigate sensitive topics, and provide thoughtful advice on interpersonal situations. When to use it: Pick GPT-4.5 when you need help with delicate communication, creative collaboration, or brainstorming sessions where you want more nuanced, human-like responses. It's particularly valuable for workplace communication, relationship advice, or any situation where tone and empathy matter. The o3 series represents OpenAI's most advanced reasoning models, with particular strength in technical and scientific tasks. What they're good for: o3 excels at complex coding projects, advanced mathematics, scientific analysis, strategic planning, and multi-step technical problems. o3-mini handles similar tasks but focuses on speed and cost-efficiency for simpler coding and math problems. When to use them: Use ChatGPT o3 for your most challenging technical work — complex software development, advanced mathematical modeling, extensive research projects, or strategic business planning. Choose o3-mini for everyday coding tasks, basic programming questions, quick prototypes, and straightforward technical problems. The newest addition to OpenAI's lineup, o4 mini is designed for users who need reasoning capabilities but prioritize speed and cost-efficiency. What it's good for: o4 mini excels at quick technical tasks, fast STEM calculations, visual reasoning with charts and data, extracting information from documents, and providing rapid summaries of scientific or technical content. When to use it: Choose ChatGPT o4 mini when you need reasoning capabilities but can't wait for the slower, more comprehensive models. It's perfect for quick math problems, rapid data analysis, fast coding help, or when you need multiple quick answers rather than one deep analysis. For everyday use: Start with GPT-4o. It handles most common tasks efficiently and works with text, images, and voice. For sensitive communication: Use GPT-4.5 when tone, empathy, and nuanced understanding matter most. For complex analysis: Choose o1 or o3 when you need thorough, step-by-step reasoning and accuracy is more important than speed. For quick technical help: Pick o4 mini when you need smart answers fast, especially for math, coding, or data analysis. Get instant access to breaking news, the hottest reviews, great deals and helpful tips. Now that you've learned how to pick the right ChatGPT model for any task, why not take a look at some of our other useful ChatGPT guides? Check out I've been using ChatGPT since its release — here's 5 tips I wish I knew sooner and ChatGPT has added a new image library — here's how to use it. And, if you want to keep your data private by opting out of training, we've got you covered.
Yahoo
23-05-2025
- Business
- Yahoo
Anthropic CEO claims AI models hallucinate less than humans
Anthropic CEO Dario Amodei believes today's AI models hallucinate, or make things up and present them as if they're true, at a lower rate than humans do, he said during a press briefing at Anthropic's first developer event, Code with Claude, in San Francisco on Thursday. Amodei said all this in the midst of a larger point he was making: that AI hallucinations are not a limitation on Anthropic's path to AGI — AI systems with human-level intelligence or better. "It really depends how you measure it, but I suspect that AI models probably hallucinate less than humans, but they hallucinate in more surprising ways," Amodei said, responding to TechCrunch's question. Anthropic's CEO is one of the most bullish leaders in the industry on the prospect of AI models achieving AGI. In a widely circulated paper he wrote last year, Amodei said he believed AGI could arrive as soon as 2026. During Thursday's press briefing, the Anthropic CEO said he was seeing steady progress to that end, noting that "the water is rising everywhere." "Everyone's always looking for these hard blocks on what [AI] can do," said Amodei. "They're nowhere to be seen. There's no such thing." Other AI leaders believe hallucination presents a large obstacle to achieving AGI. Earlier this week, Google DeepMind CEO Demis Hassabis said today's AI models have too many "holes," and get too many obvious questions wrong. For example, earlier this month, a lawyer representing Anthropic was forced to apologize in court after they used Claude to create citations in a court filing, and the AI chatbot hallucinated and got names and titles wrong. It's difficult to verify Amodei's claim, largely because most hallucination benchmarks pit AI models against each other; they don't compare models to humans. Certain techniques seem to be helping lower hallucination rates, such as giving AI models access to web search. Separately, some AI models, such as OpenAI's GPT-4.5, have notably lower hallucination rates on benchmarks compared to early generations of systems. However, there's also evidence to suggest hallucinations are actually getting worse in advanced reasoning AI models. OpenAI's o3 and o4-mini models have higher hallucination rates than OpenAI's previous-gen reasoning models, and the company doesn't really understand why. Later in the press briefing, Amodei pointed out that TV broadcasters, politicians, and humans in all types of professions make mistakes all the time. The fact that AI makes mistakes too is not a knock on its intelligence, according to Amodei. However, Anthropic's CEO acknowledged the confidence with which AI models present untrue things as facts might be a problem. In fact, Anthropic has done a fair amount of research on the tendency for AI models to deceive humans, a problem that seemed especially prevalent in the company's recently launched Claude Opus 4. Apollo Research, a safety institute given early access to test the AI model, found that an early version of Claude Opus 4 exhibited a high tendency to scheme against humans and deceive them. Apollo went as far as to suggest Anthropic shouldn't have released that early model. Anthropic said it came up with some mitigations that appeared to address the issues Apollo raised. Amodei's comments suggest that Anthropic may consider an AI model to be AGI, or equal to human-level intelligence, even if it still hallucinates. An AI that hallucinates may fall short of AGI by many people's definition, though. This article originally appeared on TechCrunch at
