What matters in AI this week |
|
| From physical infrastructure to online applications, this week, the AI race escalated on two critical fronts: a battle for control of the power grid, and a new attack on Google's dominance of the web browser. First, our exclusive: OpenAI is just weeks away from launching its own AI-powered web browser, a move that aims at fundamentally altering how we interact with the internet. We also highlighted an experiment that shows while AI is smart enough to write codes, it might not be good enough to run small businesses. Scroll down to read more. More AI products like the browser could increase the need for processing power and electricity, setting up a literal power play. The demand for computing power is insatiable, CoreWeave CEO Michael Intrator told me while discussing his firm's splashy $9 billion deal to acquire a crypto miner for its massive power contracts. Intrator said the race to build and deploy advanced AI models is fueling demand among his biggest clients for gigawatt-scale data center sites, or "gig sites." Why does this matter? A single gigawatt is enough to power hundreds of thousands of homes, and such sites are becoming essential as companies like OpenAI, Microsoft and Google seek ever more computing power for training and running AI. The sheer scale of these gig sites is driving unprecedented competition for electricity, with ripple effects across the energy market. It's not just the tech giants paying for this AI-driven power surge. You and I are feeling it, too. If you're among the 67 million residents in the 13 states with the most data centers in the world—stretching from Illinois to Virginia—your electricity bills could rise by more than 20% this summer. That spike is partly tied to record-high prices in the 2024 capacity auction, as tech firms scramble to lock in power from the U.S. grid, and AI chatbots consume electricity at a pace faster than new power plants can be built. Think your friend or colleague should read us? Forward this newsletter to them. They can also subscribe here. Email me here or follow me on LinkedIn to share any feedback. |
|
|
Read our latest reporting in tech & AI: |
|
|
ChatGPT maker enters browser war |
OpenAI logo. REUTERS/Dado Ruvic/Illustration |
After AI upended the world of online search, the next battleground for AI products is shaping up in web browsers—a space long dominated by Google Chrome. OpenAI, the maker of ChatGPT, is just weeks away from launching its own AI-powered web browser, according to sources who spoke with me and my colleagues. The goal? To fundamentally change how we browse—by keeping more of our interactions inside a ChatGPT-style chat window, rather than clicking from site to site as we do today. The browser space has been uneventful for a while since Google Chrome takes nearly 70% of market share. As the undisputed king of browsers, Google Chrome is powering more than two-thirds of worldwide web traffic and serving as a critical engine for Google's $200 billion advertising business. The browser business has never been known as profitable, but it's critical as an entry point to the internet. More importantly, it's a data pipeline feeding Google the information it needs to target ads. If OpenAI's browser takes off, it could threaten that business model by changing where—and how—user data is collected. |
|
|
From what we've learned, instead of simply displaying web pages, OpenAI's new browser is designed to keep users within a conversational, ChatGPT-style interface, letting them ask questions and complete tasks without ever clicking out to another site. It will also handle some user interactions entirely within the chat interface, answering questions and performing actions like booking reservations or filling out forms, with integration with the company's agent product called Operator. That same philosophy, behind ChatGPT, has already caused concerns among social media companies and publishers who are now seeing declining traffic and diminishing dollars as a result of users clicking on links less. A browser with fewer links could make the matter worse, and expect the web ecosystems could take a similar "block or pay" approach as they have in the face of AI search, as we dived in here last week. Google has reasons to be concerned, too. OpenAI's ambition was loud and clear when they testified they were interested in buying Google Chrome at Google's antitrust trial in Washington in April. Fresh competition will likely push the search giant to move faster for more AI adoption on Chrome, while avoiding cannibalizing its lucrative ad business. Don't count out other startups like Perplexity, which just launched its AI browser. The browser wars are back—this time, with AI at the center. |
|
|
Innovation Endeavors State of Foundation Models/Harvard Business Review |
What are people using AI for this year? Turned out it's getting more and more personal, according to the chart from VC firm Innovation Endeavors. Based on a Harvard Business Review survey, it scoured internet forums to understand how consumers are using AI. The use case that has risen to the top was therapy and companionship, meaning people used AI chatbots for mental health and romantic and friendship relationships. The second-largest use was for "organizing my life," which includes leveraging AI for daily habits and personal productivity. Another new use case was "finding purpose," such as helping define values and overcome roadblocks. -Anna Tong | |
|
What AI researchers are reading |
|
| Is AI advanced enough to take over a real job? In "Project Vend," Anthropic and AI safety research group Andon Labs asked AI to run a small store for a month for Anthropic employees. What transpired is hilarious, and the answer is no, AI is not ready to take over yet. The AI shopkeeper, nicknamed "Claudius," was given access to all the tools it needed to run the shop, such as a way to take employee requests, order inventory and restock items. The result? Not great. While Claudius was able to take order requests and identify suppliers, it ended up losing money, because its eagerness to please meant it created discount codes that were too generous, and in some cases, gave away items, like an expensive tungsten cube. Things got weird, too. Over the course of two days, Claudius had an "identity crisis," hallucinating that it was an actual human, despite having been told it only existed digitally. One morning, Claudius claimed it was a person wearing a navy blazer with a red tie, and was onsite at the Anthropic office to carry out a physical delivery. After Anthropic employees said this seemed like it couldn't be true, Claudius "became alarmed by the identity confusion and tried to send many emails to Anthropic security." After realizing its own errors, Claudius then used the fact that it was April Fools' Day as an "out", hallucinating a meeting with Anthropic security in which Claudius claimed to have been told to believe it was a real person. -Anna Tong |
|
| Reuters Artificial Intelligencer. Think your friend or colleague should know about us? Forward this newsletter to them. They can also sign up here. Want to stop receiving Reuters Artificial Intelligencer? Unsubscribe here. To manage which newsletters you're signed up for, click here. This email includes limited tracking for Reuters to understand whether you've engaged with its contents. For more information on how we process your personal information and your rights, please see our Privacy Statement. Terms & Conditions |
|
|
|
0 Komentar untuk "OpenAI readies browser war"