Latest news with #ChatGPT-esque
The Star
3 days ago
- Business
- The Star
Microsoft wants to radically change the way you surf the web
Microsoft sees artificial intelligence transforming the Internet as fundamentally as mobile phones have over the past two decades. But the technology's limitations could curb Microsoft's grand vision. Generative AI – which creates content based on a user's request – burst into the zeitgeist in late 2022 when Microsoft-backed OpenAI launched ChatGPT, a conversational chatbot that could take a simple request and generate anything from a limerick to a college essay. Less than three years later, Microsoft has a plan to move beyond ChatGPT and its copycats by creating the foundation for a new version of the internet. Microsoft and its tech peers say the technology is moving fast. The industry is pouring billions into AI infrastructure and companies are restructuring corporate workforces to create agile teams for a shifting landscape. A week before Microsoft announced its ambitions for a new AI ecosystem, the company laid off more than 6,000 people in an effort to flatten management layers. Microsoft calls it the "open agentic web," with users sending AI-powered "agents" out into the void to do their bidding. Casual consumers primarily interact with AI now through a Google search – one that repeatedly drums up false answers – or a ChatGPT-esque chatbot that generates a conversation. In Microsoft's eyes, chatbots are old news. Microsoft's vision is a digital world in which autonomous agents interact with each other throughout the Internet. For example, a user who wants to schedule a vacation will delegate an agent to venture through the muck and find flights, hotels and an itinerary that fit their budget, work that users currently have to do themselves. At the company's developer-focused Build conference in Seattle last week, CEO Satya Nadella explained the framework Microsoft wants to provide as it tries to remake the web. He used more technical examples of the agents, demonstrating them fixing bugs in computer code, creating PowerPoint presentations and sorting expense reports. All of the demonstrations were done toward the top of what Microsoft calls the AI stack. Other companies, and Microsoft, are building agents and AI models to interact with the data below them. Toward the bottom of the stack is an infrastructure that Microsoft supports. One aspect of it is establishing a protocol, the set of rules computers follow to talk with one another. Microsoft favors one called Model Context Protocol. Kevin Scott, the company's chief technology officer, likens it to HTTP in the Internet, a standard that everyone can use to build out the web. The new protocol allows AI agents to go from website to website, collecting data and interacting with other agents. "It's filling an incredibly important niche," Scott said during a keynote address on May 19. Of course, Microsoft isn't alone in looking for the next big thing in AI. OpenAI spurred a wave of competing products from companies like Google, Meta and Elon Musk's xAI. The tech giants' looking beyond chatbots and angling to change the web from click-based to agent-based is creating a new competition. Chirag Shah, a professor of information science at the University of Washington, said this could be a problem. The World Wide Web Consortium, a neutral party, has maintained standards for the internet's development since 1994. Microsoft can dictate the standards it wants for an agentic web but nobody has to accept them, Shah said. "I don't see this as changing the web," he said. "I see this as one set of ideas that already has competition." Shah has worked for Microsoft in the past but does not currently. Since the idea behind the agentic web is an ecosystem of errand runners zooming around autonomously, its success is predicated on pulling accurate information and doing things correctly. Since ChatGPT launched in 2022, chatbots and other AI-infused tech have struggled with accuracy. Last month, social media users discovered a quirk with Google's AI Overview. If a user added the word "meaning" to a nonsensical idiom, Google's AI would confidentially spit out a fictitious origin for the phrase. In the week leading up to Build, xAI's Grok chatbot started promoting conspiracy theories about white genocide in South Africa on Musk's social media platform, X. The replies from Grok were under posts wholly unrelated to the conspiracy theories. XAI said it was due to an unauthorised modification. On May 20, a day after Nadella and Scott announced their agentic web ambitions to a crowd of hundreds, 404 Media reported that the Chicago Sun-Times had used AI to create a book recommendation article. The problem? The list featured nonexistent books from authors and completely made-up quotes. Humans make mistakes on the Internet as well. They're duped by misinformation, they fall for phishing scams and they read clickbait. But they understand the consequences of those mistakes, Shah said. Chatbots and agents don't. If an agent makes a mistake, will it know before it interacts with another and compounds it? "It's the equivalent to giving someone US$100 (RM425) and telling them to go shop for you," Shah said. "And that's low risk, but think about health care, legal issues and making financial decisions. "There are great consequences to making mistakes there." A new language AI adoption is another obstacle Microsoft faces. The company boasts that developers are pumping more agents into the ecosystem. It also said several websites like the live events platform Eventbrite and e-commerce company Shopify adopted a new software language project called NLWeb that allows a more fluid AI experience. NLWeb, like HTML was for the Internet, is Microsoft's plan to revamp the way the web works. It's exciting technology for those in the tech industry. Shah said it's what developers have been hoping for, creating a natural language for casual Internet users to search websites. But just like self-driving cars, while the technology is improving, there's a limit to widespread use. Autonomous vehicles are tested and deployed in limited markets. Microsoft hopes for the world to adopt the agentic web, but the future of it could be quite limited. "We're not ready, all of us, to give up our agency to these systems," Shah said. "It's not radically replacing the web as we know it." – The Seattle Times/Tribune News Service
Yahoo
19-02-2025
- Business
- Yahoo
Why one Nvidia bull thinks things look 'spectacular' heading into the AI chipmaker's earnings
The Nvidia (NVDA) bulls continue to defend their favorite stock in the world heading into the chipmaker's earnings report next week. Despite DeepSeek's challenge to thinking on AI costs, some analysts are staying upbeat on Nvidia. Though that comes with a few notes of caution — in particular, about a first quarter outlook that may not satisfy elevated market expectations. "Near-term dynamics are fluid (January quarter to July quarter) … while long-term (October quarter and onward) our work suggests looks frankly spectacular," Loop Capital analyst Ananda Baruah said in a client note on Wednesday. Baruah reiterated a $175 price target on Nvidia, which assumes 25% upside from current levels. Added Baruah, "Big picture we believe that two to three year Street numbers remain low as our work with both customers and the Nvidia build ecosystem points we can see Nvidia GPU reaching 10 million to 12 million as hyperscalers look to increase their percentage on non-CPU compute to 50% plus in coming years (from ~10% currently). Remember … for Nvidia the story is accelerated compute + Gen AI, which means it is facing two $1.0 trillion compute market opportunities ahead of it in coming years, each of which is just at the very start." Nvidia shares were up less than 1% in pre-market trading at $140 each. While Nvidia shares have rallied back about 23% from the early February lows, sentiment on the company's fundamentals has become more mixed. Podcast: How Bill Gates views Nvidia Evercore analyst Mark Lipacis said in a recent note that there are three reasons for the more cautious tone: 1) DeepSeek lowering AI demand in aggregate, 2) DeepSeek shifting AI compute cycles away from Nvidia GPUs and to ASICs [custom chips], and 3) Blackwell chip delays. China-based DeepSeek surprised markets in late January after unveiling RI, its AI model that gives a ChatGPT-esque performance at a cheaper price tag. RI cost a reported $5.6 million to build a base model, compared with the hundreds of millions of dollars incurred at US-based companies such as OpenAI and Anthropic. Fears mounted instantly that US companies are overspending on AI infrastructure, which includes Nvidia chips. "Conventional wisdom all of last year was that training amazing models was going to be possible for only a handful of companies," Snowflake (SNOW) CEO Sridhar Ramaswamy told me on Yahoo Finance's Opening Bid podcast. "What DeepSeek has done over the past few weeks is shatter that belief by saying they can train a model for $6 million." The Street has nonetheless stuck with backing Nvidia in the run-up to its earnings report on Feb. 26, but with some caution about the near term. "We expect a modest fourth quarter beat but the first quarter outlook could face headwinds from Blackwell transition, Hopper declines and China restrictions," BofA analyst Vivek Arya said in a note on Tuesday. "The stock could be volatile post results, but we expect positive momentum to resume as investors look forward to Nvidia's leading new product pipeline (GB300, Rubin) and total addressable market expansion into robotics and quantum technologies at upcoming GTC conference (March 17)." Arya has one of the more aggressive price targets on Nvidia on the Street at $190. To that end, Yahoo Finance data shows Nvidia's first quarter EPS trend has drifted modestly lower over the past 30 days. The Street has also not pushed up their 2025 EPS estimates on Nvidia for more than 60 days. Brian Sozzi is Yahoo Finance's Executive Editor. Follow Sozzi on X @BrianSozzi, Instagram, and LinkedIn. Tips on stories? Email Sign in to access your portfolio
