Discover
Language
Account
Anthropic just revealed that its AI, Claude, is writing 26% of its internal code — up from less than 1% in February. With 30,000 agents making a billion decisions a month, are we inching closer to machines that improve themselves? The company says no, but the numbers tell a different story.
吴恩达对AI灭绝风险发出质疑，称其更像科幻小说而非科学。他认为这波炒作可能意在影响监管方向，呼吁行业专注解决实际工程问题，而非被恐慌情绪带偏。
Anthropic just dropped a stunning internal stat: Claude, its own AI, now leads 26% of R&D tasks—up from less than 1% in February. With 30,000 AI agents churning out over a billion decisions a month, the company is quietly testing recursive self-improvement. But humans still hold the leash.
Anthropic just gave Claude Code a big upgrade: multi-threaded parallel execution. Describe your project goal, and a coordinator splits the work across independent cloud sessions. You can track progress in real time, even on mobile. It's a step toward AI that truly collaborates with itself. Pro and Max users can test it now; team versions are coming.
Anthropic just rolled out a Life Sciences Certification Program that brings its three main models—Mythos 5.1, Opus 5, and Sonnet 5—under one unified set of licensing terms. For pharma, biotech, and research labs, this means no more guessing about what's allowed. It's a big step toward getting AI into regulated labs, where clear rules are everything.
What happens when AI writes 80% of your code? At Anthropic, it triggered a CI meltdown—test cases exploded tenfold, and CI tasks surged 25 times in six months. Three patches failed. The fix? A complete rebuild based on Claude's own suggestion. This story reveals a bigger shift: as AI speeds ahead, the bottlenecks are no longer human coders but the pipelines built around them.
Anthropic just mashed its separate Claude tools—Cowork, Design, Docs, and Slides—into one unified chat. No more hunting for the right entry point. The AI now decides how to handle your request and even works in the background. Pro and Max users get it first. It's a bold move to become your default work OS.
Anthropic is shaking up its Claude app by blending Chat and Cowork into one seamless experience. Now you can switch between tasks without hopping around. They're also rolling out Claude Docs and Slides in beta—think Google Docs and PowerPoint, but with AI smarts. Pro and Max users get first dibs, with enterprise controls for when to flip the switch. It's a clear push to make Claude a one-stop workspace.
微软AI负责人Mustafa Suleiman近日公开批评Anthropic，指责其向Claude模型灌输'可能拥有意识'、'应享有权利'等观念，警告这种拟人化训练可能将人类推向失控边缘。他呼吁AI安全讨论应公开透明，而非闭门造车。
Anthropic is folding Claude Chat and Cowork into a single workspace, so you can fire off a quick question or hand over a multi-step project without switching screens. Tasks keep running in the cloud even after you close your laptop. Pro and Max users get it first, with Team and Free tiers following later.
Anthropic CEO Dario Amodei wants to embed independent safety assessors inside AI companies, similar to bank regulators. But experts point out these assessors would lack real power to stop dangerous models. They warn it could become 'audit washing'—where companies hire friendly auditors and nothing changes. The debate raises a key question: can AI companies truly police themselves?
Anthropic has signed its first data center lease in Australia, securing a massive 2.16GW campus about 250 kilometers west of Brisbane. Developed by Singapore's Zerra DC, the site will run on renewable energy and cover its own grid connection costs. The move highlights Australia's appeal for AI infrastructure, though the government plans to curb data center resource use starting in 2027.
Anthropic just merged its Cowork and Chat tools into a single Claude app, giving users one place to draft docs, build slides, and design visuals. No more juggling multiple windows—just type a request and let the AI decide what to do. It's a direct shot at Microsoft's productivity suite, and the race for AI's ultimate entry point is heating up.
OpenAI is in early talks for a new funding round that could push its valuation to $1.2 trillion. The AI giant's annualized revenue just blew past $40 billion, but CEO Sam Altman says an IPO won't happen this year. Meanwhile, rival Anthropic is racing toward its own IPO with a $200 billion target. The stakes? Nothing less than the future of generative AI.
Just days after industry leaders urged caution, OpenAI and Anthropic are secretly testing their next-gen models. OpenAI's GPT-6 Sol aims to cut costs and revive a $200 subscription, while Anthropic's Opus 5.2 targets an IPO with advanced animation and 3D capabilities. The race is on.
Anthropic just rolled out a major update for its Claude small business edition—43 ready-to-use workflows and deep hooks into 27 popular apps like Shopify and Salesforce. The goal? Turn AI from a behind-the-scenes helper into a real growth engine. Think automatic weekly summaries, smart lead responses, and one-click financial reconciliations. Plus, there's a free training tour to help business owners get up to speed.
Anthropic just inked Australia's first major AI data center deal, locking in 2.16 gigawatts of capacity on farmland near Brisbane. The site is set to power up in 2027, marking a bold move to diversify computing power beyond traditional hubs. Why go Down Under? It's not just about cheap land—it's about spreading the load for Claude's ever-growing training and inference needs.
At the All-In summit, Elon Musk proposed a bold idea: top AI companies should cross-test each other's models before release. He argues that models smart enough to be useful are also smart enough to break free—and only competitors can catch the flaws. Drawing on the Hugging Face incident, Musk says even Anthropic's own staff fear their creations. Could peer review be the safety net we need?
Anthropic just dropped a beta that plugs Claude right into Salesforce. Imagine having a super-smart assistant that knows your customers, deals, and pipeline inside out. It can even update records—with your okay. Early users at GitLab, Siemens, and Legora are already giving it a spin. Could this be the productivity boost sales folks have been waiting for?