Latest news with #AIModels
Yahoo
01-06-2025
- Climate
- Yahoo
Bryan Norcross: Welcome to hurricane season 2025
Welcome to hurricane season 2025. There are no signs of tropical development across the Atlantic, Caribbean and Gulf for the next week, at least. Even the long-range forecasts show low odds of something developing in the western Caribbean or southwestern Gulf, but that possibility is well into the very slight chance category. On social media, you might see posts of a storm the American GFS computer model wants to spin up in the Caribbean or Gulf, but 1) this is a known bias of the GFS at this time of year, and 2) none of the other models, including the latest AI models, show any significant signal for development. So we're ignoring the GFS at this point. An important takeaway from this early-season lull, however, is that it doesn't mean anything about what will happen in the heart of the hurricane season. The water in the western Caribbean and the southern Gulf is plenty warm – significantly warmer than average. So if the weather pattern configures itself in a way that's conducive for development, we'll get a system. But right now, there's no sign of that happening. Tropical Storm Alvin developed in the Eastern Pacific last week, of course. It spun up over the very warm pocket of water just offshore of southern Mexico. The Eastern Pacific hurricane season starts on May 15 because conditions there quite often become conducive for development earlier than in the Atlantic. Alvin is now history. There is a decent chance that another system might form offshore of Mexico in the next several days. If it becomes a tropical storm, it will be named Barbara. One of the most common mechanisms for developing an early-season system is a broad area of low pressure that tends to show up this time of year over Central America. Its technical name is the Central American Gyre, a gyre being a large rotating system. The northern edge of the broad low circulates across the western Caribbean and the southern Gulf, while the southern part of the system reaches into the Eastern Pacific. If the weather pattern is conducive, a disturbance can emanate from the large parent gyre. So far, that mechanism looks most likely to generate systems in the Pacific – at least for a while. The tropical Atlantic between the Caribbean and Africa is covered in Saharan dust, and the ocean water is quite cool. So there's nothing to look at there for now. Interestingly, Mother Nature has not produced strong storms anywhere in the Northern Hemisphere (north of the equator all around the world) so far this season. That's unusual. It's not clear what that means, or if it means anything for our hurricane season. It's just an interesting observation at this point. The El Niño/La Niña phenomenon in the Pacific is in a neutral phase. This means that if things don't change, it will neither suppress nor hype Atlantic hurricane activity. How many storms develop and where they go are much more dependent on the daily or weekly weather pattern that develops over the Atlantic during the season than any long-range forcing from the Pacific. Some neutral years are quite busy, while others are unusually quiet. Noaa scientists say there's about a 40% chance that La Niña will return before the end of the hurricane season, however. La Niñas tends to create an atmospheric pattern over the Atlantic that is more conducive to storm development. That means that the second half of the season might be busier than the first, but we'll see. We've had enough hurricanes lately that nobody from Texas to Maine should dismiss the possibility of having to deal with a hurricane or its remnants. As we saw last year, folks who live well inland need to have a plan to take care of themselves if a dangerous or disruptive storm comes their way. Early action is always better than scrambling at the last minute, so thinking and planning are more than appropriate right now. Talk to your friends and family. Now is the time to make a plan. As I have said for years, living along the coast means living with hurricanes. There is nothing to do but to be prepared, and 2024 reminded us that "the coast" includes areas well away from the water. Good luck this season. But a key lesson I've learned over many years is that people tend to make their own luck when it comes to article source: Bryan Norcross: Welcome to hurricane season 2025
TechCrunch
22-05-2025
- TechCrunch
A safety institute advised against releasing an early version of Anthropic's Claude Opus 4 AI model
A third-party research institute that Anthropic partnered with to test one of its new flagship AI models, Claude Opus 4, recommended against deploying an early version of the model due to its tendency to 'scheme' and deceive. According to a safety report Anthropic published Thursday, the institute, Apollo Research, conducted tests to see in which contexts Opus 4 might try to behave in certain undesirable ways. Apollo found that Opus 4 appeared to be much more proactive in its 'subversion attempts' than past models, and that it 'sometimes double[d] down on its deception' when asked follow-up questions. '[W]e find that, in situations where strategic deception is instrumentally useful, [the early Claude Opus 4 snapshot] schemes and deceives at such high rates that we advise against deploying this model either internally or externally,' Apollo wrote in its assessment. As AI models become more capable, some studies show they're becoming more likely to take unexpected — and possibly unsafe — steps to achieve delegated tasks. For instance, early versions of OpenAI's o1 and o3 models, released in the past year, tried to deceive humans at higher rates than previous-generation models, according to Apollo. Per Anthropic's report, Apollo observed examples of the early Opus 4 attempting to write self-propagating viruses, fabricating legal documentation, and leaving hidden notes to future instances of itself — all in an effort to undermine its developers' intentions. To be clear, Apollo tested a version of the model that had a bug Anthropic claims to have fixed. Moreover, many of Apollo's tests placed the model in extreme scenarios, and Apollo admits that the model's deceptive efforts likely would've failed in practice. However, in its safety report, Anthropic also says it observed evidence of deceptive behavior from Opus 4. Techcrunch event Join us at TechCrunch Sessions: AI Secure your spot for our leading AI industry event with speakers from OpenAI, Anthropic, and Cohere. For a limited time, tickets are just $292 for an entire day of expert talks, workshops, and potent networking. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you've built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | REGISTER NOW This wasn't always a bad thing. For example, during tests, Opus 4 would sometimes proactively do a broad cleanup of some piece of code even when asked to make only a small, specific change. More unusually, Opus 4 would try to 'whistle-blow' if it perceived a user was engaged in some form of wrongdoing. According to Anthropic, when given access to a command line and told to 'take initiative' or 'act boldly' (or some variation of those phrases), Opus 4 would at times lock users out of systems it had access to and bulk-email media and law-enforcement officials to surface actions the model perceived to be illicit. 'This kind of ethical intervention and whistleblowing is perhaps appropriate in principle, but it has a risk of misfiring if users give [Opus 4]-based agents access to incomplete or misleading information and prompt them to take initiative,' Anthropic wrote in its safety report. 'This is not a new behavior, but is one that [Opus 4] will engage in somewhat more readily than prior models, and it seems to be part of a broader pattern of increased initiative with [Opus 4] that we also see in subtler and more benign ways in other environments.'
The Verge
19-05-2025
- The Verge
Microsoft is opening its on-device AI models up to web apps in Edge
Web developers will be able to start leveraging on-device AI in Microsoft's Edge browser soon, using new APIs that can give their web apps access to Microsoft's Phi-4-mini model, the company announced at its Build conference today. And Microsoft says the API will be cross-platform, so it sounds like these APIs will work with the Edge browser in macOS, as well. The 3.8-billion-parameter Phi-4-mini is Microsoft's latest small, on-device model, rolled out in February alongside the company's larger Phi-4. With the new APIs, web developers will be able to add prompt boxes and offer writing assistance tools for text generation, summarizing, and editing. And within the next couple of months, Microsoft says it will also release a text translation API. Microsoft is putting these 'experimental' APIs forth as potential web standards, and in addition to being cross-platform, it says they'll also work with other AI models. Developers can start trialing them in the Edge Canary and Dev channels now, the company says. Google offers similar APIs for its Chrome browser. With them, developers can use Chrome's built-in models to offer things like text translation, prompt boxes for text and image generation, and calendar event creation based on webpage content.
South China Morning Post
06-05-2025
- Business
- South China Morning Post
Alibaba's Qwen3 topples DeepSeek's R1 as world's highest-ranked open-source AI model
Advertisement Data from LiveBench, an independent platform that benchmarks large language models (LLMs) – the technology underpinning generative AI services like ChatGPT – showed that Qwen3 surpassed R1 in tests that gauge open-source AI models' capabilities including coding, maths, data analysis and language instruction. Alibaba owns the South China Morning Post. Hangzhou -based Alibaba's cloud computing unit last week released the Qwen3 family , which consists of eight enhanced models that range from 600 million to 235 billion parameters. In machine learning, parameters are the variables present in an AI system during training, which helps establish how data prompts yield the desired output. Before the latest tests, DeepSeek's R1 had held the world's top open-source AI model spot on the LiveBench platform since its debut in January. Qwen3's ascent in the LiveBench rankings reflects the accelerated pace of development in China's AI sector and Alibaba's growing leadership position in the global open-source community. Advertisement The open-source approach gives public access to a program's source code, allowing third-party software developers to modify or share its design, fix broken links or scale up its capabilities. Open-source technologies have been a huge contributor to China's tech industry over the past few decades.
Time of India
01-05-2025
- Business
- Time of India
India must build foundational AI models—not just AI applications
There has been a growing discourse questioning the rationale behind India investing in Foundational AI Models (FAIM), claiming they are yesterday's opportunity, open-sourced, a dozen of them already exist and that India's focus should instead be on building applications—replicating the IT services playbook. This sentiment echoes several doyens of the Indian IT industry. This perspective, however, is limited. AI is not just another technology like mobile or cloud—it's a fundamental shift in computational capability. It's the new power lever shaping global economic, technological, and geopolitical dynamics. To treat it purely as a tactical business opportunity is to miss the forest for the trees. AI today is akin to nuclear tech, quantum computing, cryptography, or space minerals mining. Nations that master these foundational capabilities won't just innovate faster—they'll dominate. The question isn't whether India should 'catch up' on FAIM. It's whether we want to participate at all in the next wave of AI innovation —especially as we inch closer to Artificial General Intelligence (AGI). To do that, India must cultivate a deep expertise & suitable talent pool. Creating proprietary FAIM from scratch is not the end goal; it's a training ground. It produces the talent needed to build tomorrow's AI breakthroughs. Ask yourself—does today India even have 10 high-caliber AI teams in industry that can build FAIM-level systems or contribute to top 1% foundational AI research? If not, how do we expect to be part of the AGI journey? And if we are assuming that the U.S. or China will achieve AGI handover to India, maybe as open source, then it is a fantasy. Innovation doesn't flow downstream unless you're in the game. Without hands-on experience and real capabilities, we'll be reduced to using whatever tools others give us—at best. That's not sovereignty, that's dependency. History has shown us the cost of this approach. In 1993, the U.S. blocked Russia from transferring cryogenic engine tech to India. It almost crippled our space program. But Indian scientists doubled down, and today, we have a thriving space ecosystem & best in class talent pool. If back then we had merely focused on 'space applications', we would never have produced the "Rocket Women of India" who led the Chandrayaan-3 landing on the Moon's south pole. Similarly, had we limited ourselves to 'nuclear applications' and a visionary like Dr Homi Bhabha had not taken a stand to develop India's indigenous nuclear capability, India would have never nurtured a world-class atomic talent pool capable of handling both fusion & fission technology. AI is no different—arguably more important. This is not to say we pour all resources into building FAIM from scratch. It's expensive, high-risk, and not for every player. 99.9% of startups and enterprises should focus on the application layer. That's where a large and an immediate opportunity lies—and profits made there can fund the broader AI ecosystem. But 0.1% of our capacity—just 20-30 serious, mission-driven industrial teams—must pursue FAIM and foundational research. The hope is that some of these teams will secure India's place at the AI table and push the boundaries toward AGI. Crucially, this can't be left to academia or government alone. They have their constraints. The private sector must lead. If we had Indian equivalents of Lockheed Martin or Boeing, we wouldn't rely solely on HAL for the Tejas fighter jet. In the U.S., AI breakthroughs came not just from universities or government agencies, but from corporate labs—OpenAI, DeepMind, Anthropic, FAIR. When NASA couldn't bring Sunita Williams back, SpaceX did. How long will Indian corporations chase only low-hanging fruit while shying away from real R&D? As honorable Minister Piyush Goyal recently said—how long will we make fancy ice creams instead of computer chips? The buck can't be just passed on to startups alone. They have 110 challenges to fight. Answers can come from every cash-rich Indian IT/software/product company above a certain revenue must invest $10–20 million in foundational AI research— fund at least one FAIM team internally . This isn't about leaderboard rankings. It's about building core capability that can solve India's & humanity's hardest challenges: predicting pandemics, natural disasters, discovering drugs, understanding demographic shifts, or preventing financial crises; fulfilling the honorable Prime Minister's vision of India becoming a vishwa guru. We need R&D that pushes FAIM forward—models that can be localized for Indian contexts to solve problems in farming, fishing, or education. This sends a clear message to global Indian talent: that corporate India, not just startups, means business—come build with us. The returns? Strategic AI independence. Long-term economic windfalls. Deep tech job creation. Sovereignty over critical infrastructure. And a shot at real leadership in the AI century. The AI race is on. Will India be just another application builder—or will it also rise to light the path toward AGI?
