π€ AI Just Disproved an 87-Year-Old Math Problem β And That's Just the Start
Welcome to Daily Inference, your daily briefing on the world of artificial intelligence. I'm your host, and today is Tuesday, July 21st, 2026. We've got a packed episode covering everything from AI cracking an 87-year-old math mystery, to the US military running out of AI tokens, to a brewing geopolitical war over open-source models. Let's dive in.
But first, a quick word from our sponsor. Today's episode is brought to you by 60sec.site β the AI-powered tool that lets you build a stunning website in, you guessed it, 60 seconds. Whether you're launching a project, a portfolio, or a business, 60sec.site makes it ridiculously fast and easy. Check them out at 60sec.site.
Alright, let's start with something that genuinely made our jaws drop. Anthropic's Claude has apparently disproved an 87-year-old mathematical conjecture. Let that sink in for a moment. A problem that mathematicians have been wrestling with since the 1930s, and an AI model cracks it. We don't yet have the full technical breakdown published, but this puts Claude squarely in the conversation about AI as a genuine scientific collaborator β not just a tool for summarizing emails or writing cover letters. This is the kind of result that the research community has been both anticipating and dreading. It raises a fascinating question: if AI can disprove century-old math, what does that mean for the future of academic disciplines that rely on human ingenuity as their core value proposition?
Speaking of Anthropic, the company just had a major legal moment. A one-point-five billion dollar copyright settlement was officially approved, making it one of the largest payouts in AI litigation history. Now, the important context here is what this settlement does NOT do β it doesn't resolve the fundamental question of whether using copyrighted material to train AI models is legal or not. That debate is very much still alive, and in fact Sony Music just filed a brand new lawsuit against AI music generator Udio, listing over thirty thousand songs allegedly used without permission β tracks from Elvis Presley, BeyoncΓ©, Harry Styles, and many more. Sony claims that list is just the tip of the iceberg. So while Anthropic's settlement closes one chapter, the broader copyright battle across text, music, and imagery is intensifying. The legal framework for how AI gets trained is being written in courtrooms right now, and the stakes are enormous.
Now let's talk geopolitics, because this story connects several threads we've been watching. China is delivering a serious one-two punch to American AI dominance. Beijing-based Moonshot AI dropped Kimi K3, a model the company claims beats nearly every US system in its own testing. And Alibaba followed up with a preview of Qwen3.8-Max β a jaw-dropping two-point-four trillion parameter multimodal model. For context, that's a massive mixture-of-experts architecture, and Alibaba is calling it second only to top frontier models. Now, to be fair, we don't have independent benchmark verification yet, and skepticism is warranted. But even the preview is rattling cages. It's no coincidence that OpenAI has been vocal about its fears of Chinese open-weight models, and Trump's AI advisory circle is apparently in open warfare over how to respond. In fact, the director role at the Center for AI Standards and Innovation has now become a revolving door β the latest AI czar has already resigned. That's a significant governance vacuum at a time when the geopolitical stakes couldn't be higher.
On the infrastructure side of things, two interconnected stories paint a stark picture of AI's physical footprint. The UK water industry has issued a blunt warning: the country simply doesn't have enough water to support its planned AI datacentre expansion. These facilities consume enormous quantities of water for cooling β through towers, chillers, and humidification systems. Meanwhile, across the Atlantic, the US Army sent internal emails warning soldiers they were burning through their AI token allocations at an alarming rate and needed to dial back usage. Think about that. The military β an organization with virtually unlimited compute budgets compared to most β is hitting usage limits. It tells you something profound about just how hungry modern AI systems are for resources. Whether it's water, electricity, or tokens, the infrastructure ceiling is becoming a real constraint on AI ambition, and that gap between aspiration and physical reality is going to define a lot of policy decisions in the months ahead.
Finally, let's look at what's happening on the job front β because it's a story with real human consequences. Australia's Nine Entertainment, which owns the Sydney Morning Herald and The Age, just announced thirty newsroom job cuts, explicitly citing AI disruption as the driver. And new research from MIT Technology Review suggests AI hiring tools are actually more prone to forming biases than human recruiters β potentially penalizing candidates in ways that are both unfair and legally concerning. So we have a double bind: AI is being used to cut the jobs of humans reviewing content, while simultaneously the AI tools screening job applicants may be doing so unfairly. It's a complicated picture, and it's playing out in newsrooms, offices, and hiring pipelines around the world right now.
On a brighter note for developers β NVIDIA released Cosmos 3 Edge, a four-billion-parameter world model designed to run directly on robots and edge devices, enabling real-time reasoning without needing to ping a cloud server. And Meta open-sourced its internal design system, Astryx, which it's been using across more than thirteen thousand apps for eight years. It ships with over a hundred and fifty accessible components and is built to be agent-friendly. Both releases signal a clear trend: AI is moving closer to the edge, closer to the physical world, and closer to being embedded in the tools developers use every day.
That's your Daily Inference for July 21st, 2026. The throughline today? AI is cracking math problems, consuming resources at staggering scale, reshaping industries, and becoming a full-blown geopolitical flashpoint β all at the same time. It's a lot to process, which is exactly why we're here every day.
Don't forget to visit dailyinference.com for our daily AI newsletter β everything we cover here, plus deeper analysis, delivered straight to your inbox. And once again, huge thanks to today's sponsor, 60sec.site. Build your next website in sixty seconds flat. We'll see you tomorrow on Daily Inference.
