Someone threw a Molotov cocktail at OpenAI CEO Sam Altman’s home at 3:45 a.m. In response, Altman published a personal blog post admitting past mistakes and comparing the AI industry’s power struggles to the “Ring of Power.”
Google’s new open-source model, Gemma 4, processes text, images, and audio completely on-device. Using agent skills, the AI can independently tap into tools like Wikipedia or interactive maps; no cloud required.
AI startup Overworld has released Waypoint-1.5, an update to its real-time world simulation system that generates interactive 3D worlds on consumer hardware. The software now runs on Mac and Windows for the first time, with two model tiers: 720p at 60 frames per second for high-performance systems, and 360p for a broader range of gaming PCs with NVIDIA RTX graphics cards and eventually Apple Silicon.
Compared to the earlier Waypoint 1.0 and 1.1 releases, the new version delivers noticeably better visual quality, improved efficiency, and stronger system performance, all while being half the size. Overworld says the model was trained on roughly 100 times more data than the original version.
Users can install the software locally through the Biome runtime environment or try it out via browser streaming at Overworld.stream. More details are available at over.world.
Anthropic has added a feature called "Ultraplan" to Claude Code that moves the planning phase of programming tasks to the cloud. Developers start a planning job in the terminal, Claude works out the plan on the Claude Code web interface, and the terminal stays free for other work meanwhile.
The documentation on code.claude.com lists three differences compared to local planning: users can leave comments on individual sections of a plan rather than responding to the whole thing at once, planning runs in the background in the cloud, and the finished plan can be executed either in the browser or back in the terminal. The browser interface also supports inline comments, emoji reactions, and revision requests.
Using Ultraplan requires a Claude Code web account, a GitHub repository, and at least version 2.1.91 of Claude Code. It doesn't work with Amazon Bedrock, Google Cloud Vertex AI, or Microsoft Foundry. Anthropic employee Thariq noted on X that Ultraplan consumes roughly the same number of tokens as the previous plan mode. The feature is currently available as a preview for anyone who has activated Claude Code on the web.
Deepmind CEO Demis Hassabis compares the arrival of AGI to ten times the industrial revolution in a tenth of the time. "I sometimes quantify AGI as 10 times the industrial revolution at 10 times the speed. So unfolding over a decade instead of a century," Hassabis says in the 20VC podcast. He sees a "very good chance of it being within the next 5 years," an assessment that hasn't changed since 2010, when co-founder Shane Legg predicted it would take 20 years: "I think we're pretty much on track."
Getting there still requires several major advances, including continuous learning, long-term planning, better memory architectures, and greater consistency. Hassabis describes current systems as "jagged intelligences," "really amazing at certain things when you pose the question in a certain way, but if you pose a question in a slightly different way they can still fail at quite elementary things." Scaling continues to deliver results, "although they're a bit less than they were at the start of all of this scaling."
Hassabis also points to a growing perception gap. "Today and in the next year things are a bit overhyped in AI," he says. But looking further out, "it's still very underappreciated how revolutionary this is going to be in the time scale of about 10 years."
According to CIA Deputy Director Michael Ellis, the agency recently produced its first fully autonomous intelligence report using AI, Politico reports. Over the next few years, AI assistants will be integrated into all of the agency's analysis platforms. These tools are meant to help analysts with tasks like drafting assessments, verifying findings, and identifying trends.
Ellis stressed that humans will continue to make the important decisions. The CIA tested 300 AI projects over the past year, covering areas like data processing and language translation. The agency's expanded Center for Cyber Intelligence, which oversees the CIA's covert hacking operations, is also set to make greater use of AI and emerging technologies.
Ellis also took an indirect shot at Anthropic, saying the CIA won't let private companies dictate how it uses their technology. Anthropic is currently in a dispute with the Pentagon after the company tried to contractually restrict its models from being used for lethal strikes and mass surveillance. The Pentagon has since classified Anthropic as a supply chain risk. Ellis also warned that China has made significant technological gains.
Coreweave has signed a multi-year cloud deal with AI startup Anthropic. The agreement will provide Anthropic with compute capacity for its Claude model family starting later this year. Financial details were not disclosed. Coreweave's stock rose more than 5 percent in premarket trading. The buildout will happen in phases, with the option to expand later.
For Coreweave, the Anthropic deal is part of a string of major contracts: last year, the company signed an $11.9 billion deal with OpenAI, followed by a $6.3 billion order with Nvidia in September, and just the day before, an expanded $21 billion deal with Meta. The Anthropic contract helps Coreweave diversify its revenue - until now, around 67 percent of its income came from Microsoft. Coreweave's stock is up about 29 percent year to date.
OpenAI is pitching investors on the idea that its early infrastructure buildout gives it a decisive advantage over Anthropic. Meanwhile, the company is pausing its UK data center project, and Anthropic is exploring custom AI chips.
