Welcome to Daily Inference, your daily briefing on everything happening at the frontier of artificial intelligence. I'm your host, and today is packed — Anthropic is on an absolute tear, OpenAI is reshuffling the deck, Netflix just dropped a bombshell open-source tool, and AI is literally writing its own algorithms. Let's get into it.

But first, a quick word from our sponsor. If you've ever wanted to build a website but didn't want to spend weeks designing or coding it, check out 60sec.site. It's an AI-powered tool that helps you create a beautiful, functional website in under a minute. Seriously. Visit 60sec.site and see for yourself.

Alright, let's start with the story that caught every filmmaker's eye this week. Netflix's AI research team has open-sourced a model called VOID — and the name is fitting, because it erases objects from video footage like they never existed. Now, you might think, big deal, photo editing apps have done object removal for years. But video is a completely different beast. When you pull an object out of a static image, you're filling in pixels. When you pull something out of video, you're dealing with motion, lighting shifts, shadows, and physical interactions that change frame by frame. Remove a person carrying something, and suddenly that object is floating in mid-air, defying gravity. Hollywood VFX teams have historically spent weeks correcting exactly that kind of problem. VOID tackles the physics of the scene, not just the pixels. The fact that Netflix is releasing this openly rather than keeping it proprietary is a significant moment for the creative and research communities. Tools that once required a studio budget and a team of specialists are now being handed to anyone with a laptop.

Now, shifting from creative AI to research AI — Google DeepMind published something genuinely mind-bending this week. They've developed a system called AlphaEvolve, which is essentially an AI that rewrites its own game theory algorithms. Think about the domain it's operating in: multi-agent reinforcement learning in imperfect-information environments — like poker, where players act in sequence without knowing each other's private cards. Designing optimal algorithms for this has always required human researchers with deep intuition, painstakingly testing different mathematical approaches through trial and error. AlphaEvolve replaces that human-driven iteration with an LLM-powered evolutionary coding agent — and not only does it generate new algorithmic approaches, it outperformed the expert-designed versions. This is a striking example of AI accelerating AI research itself. When you combine this with the open-source reasoning models we're seeing emerge, like Arcee AI's Trinity Large Thinking released under Apache 2.0 this week, you start to see a future where AI systems are actively contributing to their own next generation.

Now let's talk about Anthropic, because they had an exceptionally busy week — and it's worth connecting the dots. On the market side, Rainmaker Securities president Glen Anderson is calling Anthropic the hottest trade in the private secondary markets right now, with OpenAI actually losing ground among investors. Then came the news that Anthropic acquired a stealth biotech AI startup called Coefficient Bio in a four-hundred-million-dollar stock deal. Biotech is a massive swing for a company primarily known for building conversational AI, and it signals that Anthropic is thinking far beyond chatbots. Meanwhile, Anthropic also launched a new political action committee ahead of the midterm elections, positioning itself to back candidates aligned with its policy agenda. And on the product side, the company made waves by essentially blocking third-party tool OpenClaw from using Claude subscription limits — starting April fourth, users who want to connect OpenClaw to Claude now have to pay separately on a pay-as-you-go basis. Notably, OpenClaw's creator Peter Steinberger now works at OpenAI, which adds a competitive dimension to what might otherwise look like a routine policy change. Anthropic is clearly pushing users toward its own ecosystem tools like Claude Cowork. So you've got Anthropic expanding into biotech, entering politics, and tightening its platform control — all in the same week. That's a company operating in multiple dimensions simultaneously.

Over at OpenAI, the leadership picture got complicated. Fidji Simo, the CEO of AGI deployment, announced she's stepping away on medical leave for several weeks due to a neuroimmune condition. While she's out, OpenAI president Greg Brockman takes over product responsibilities, including the company's super app ambitions. The CMO, Kate Rouch, is also stepping back to focus on cancer treatment, with plans to return when her health allows. COO Brad Lightcap, meanwhile, is taking on a new undefined role focused on special projects. That's three major leadership moves in one news cycle. And then there's the TBPN acquisition — OpenAI bought a Silicon Valley tech talk show that broadcasts live for three hours every weekday, featuring guests like Sam Altman himself alongside executives from Meta, Microsoft, and major venture capital firms. OpenAI's chief of strategy framed this as helping the company engage with the public, but let's be real — this is a narrative play. With a lawsuit against Elon Musk heading to trial, and growing public skepticism about AI's direction, buying a media platform is a calculated move to shape the conversation.

Finally, let's zoom out to the infrastructure layer, because the AI energy story is getting harder to ignore. Google confirmed it's partnering with Crusoe Energy to power a new Texas data center campus called Goodnight through a natural gas plant that would emit four and a half million tons of carbon dioxide per year — more than the entire city of San Francisco produces annually. Meta, Microsoft, and Google are all making similar bets on fossil fuel infrastructure to keep up with AI compute demands. This is a direct contradiction of the clean energy pledges these companies made just a few years ago. And here's where it connects to something wild — SpaceX has filed with the FCC to launch up to one million data centers into Earth's orbit. MIT Technology Review broke down just how far-fetched that actually is from a current technology standpoint. But the fact that it's being seriously discussed tells you everything about how desperate the search for scalable AI infrastructure has become. The energy question isn't going away, and it's going to shape AI development in ways that go far beyond benchmark scores.

That's your Daily Inference for today. We covered Netflix's physics-aware video editing model VOID, Google DeepMind's self-rewriting game theory AI, Anthropic's wild multi-front expansion week, OpenAI's leadership shuffle and media acquisition, and the growing tension between AI's energy appetite and climate commitments. Want to stay ahead of all of this every single day? Head to dailyinference.com and subscribe to our newsletter — it lands in your inbox every morning before you've had your first coffee. And again, if you need a website built fast, let AI do the heavy lifting at 60sec.site. We'll be back tomorrow with more from the frontier. Stay curious.