AI breaks Erdős geometry conjecture & Macrophages seen eating melanoma cells - News (May 23, 2026)

We’ll start with that surprising mathematics headline. OpenAI says an experimental AI chatbot has disproved a conjecture tied to the “unit-distance problem,” a classic question about how to place many points on a plane so you get as many pairs as possible at exactly one unit apart. Paul Erdős предложил a construction back in 1946 and essentially dared the world to do better. According to OpenAI, its system found a configuration that beats Erdős’s long-standing benchmark. What makes this more than a flashy claim is the verification: OpenAI says independent mathematicians—people not affiliated with the company—reviewed the work and confirmed the improvement. Details are still incomplete in public, though. OpenAI hasn’t named the reasoning model, and it hasn’t fully released what it describes as a 125-page chain of reasoning. Even so, researchers quoted by Nature are calling it a striking moment—potentially one of the first times an AI has autonomously produced a genuinely new, mathematically meaningful result that humans can check.

Staying with science, immunology researchers in Australia have captured something that’s both visually compelling and clinically intriguing: macrophages—immune cells better known for cleanup duty—actively attacking live melanoma cells. Using high-resolution imaging in mice, a team from the Garvan Institute and UNSW Sydney filmed a particular macrophage subset, marked by a protein called CD169, patrolling the edges of tumors and engulfing cancer cells in real time. The “why it matters” here is twofold. First, when the researchers removed these CD169-positive macrophages, tumors grew larger—suggesting the cells aren’t just bystanders, they’re protective. Second, this anti-cancer activity appeared to work independently of T cells and B cells, which are the usual stars of modern immunotherapy. If this holds up in further studies, it could point toward treatments that help patients whose tumors don’t respond well to today’s T-cell-focused checkpoint drugs.

Another story where biology and AI are meeting in the middle: researchers at the University of Washington’s Institute for Protein Design, working with startup Skape Bio, say they’ve designed tiny “miniproteins” that can switch major signaling receptors on or off. These are GPCRs—G protein-coupled receptors—and they’re among the most important drug targets in medicine. The catch with GPCRs is that they’re notoriously hard to control precisely, because their binding sites can be deep and flexible, and small molecules don’t always nudge them into the exact state you want. The new approach uses AI-guided design to create very small proteins that fit into those pockets and stabilize either an active or inactive state. The team reports structural evidence that several designs match what they intended, and one mouse test suggested a designed miniprotein performed about as well as an existing drug, with fewer side effects. If this scales, it’s a potential new route for drugs where traditional discovery has hit a wall—less guesswork, more tailored control.

In regenerative medicine, researchers are also pushing organoids—mini organs grown from stem cells—toward something closer to real, functional tissue. A newly reported “confined culture system” uses a 3D-printed scaffold to guide thousands of gut spheroids to fuse into longer, tube-like tissue rather than staying as tiny spheres. The headline result is size and maturity: these constructs could be transplanted earlier and engrafted more efficiently, then grow into centimeter-scale grafts with more realistic intestinal architecture. But the standout detail is nerves. The engineered tissues developed a human-origin enteric nervous system—neurons and supporting cells—without the researchers having to add external nerve precursor cells. And those nerves weren’t just decorative: the tissue showed nerve-dependent, electrically triggered smooth-muscle contractions similar to adult intestine. That’s a big step for disease modeling, and it inches engineered gut grafts closer to being useful for patients with severe intestinal failure.

Now to the weight-loss drug race, where expectations keep climbing. Eli Lilly says a large phase 3 trial of its experimental weekly injection retatrutide produced greater weight loss than current leading obesity drugs. In a study of more than two thousand adults without diabetes, the highest dose group lost on average about 28% of body weight over 80 weeks, and nearly half lost at least 30%. Retatrutide targets multiple hormone pathways—more than many current drugs—which could explain the stronger effect. But the familiar trade-off remains: side effects were common, especially gastrointestinal issues like nausea and diarrhea, and they increased with higher doses. The big “why it matters” is that, if the full data holds up through peer review and regulators sign off, it could reset what patients and doctors consider achievable with medication—while also forcing a sharper conversation about tolerability and long-term safety.

Related to that, a large international review looking across more than 90,000 participants adds to the case that GLP-1 drugs aren’t just about weight and blood sugar. The meta-analysis pooled major cardiovascular outcome trials and found GLP-1 receptor agonists reduced the risk of major adverse cardiovascular events—heart attack, stroke, or cardiovascular death—by about 13% compared with placebo. Notably, the benefits showed up whether or not people had diabetes, with the biggest gains in those already at high cardiovascular risk, including people with obesity or established heart disease. In plain terms: these medicines increasingly look like heart-protection drugs as well as metabolic drugs—an important shift, because cardiovascular disease remains a leading cause of death worldwide.

Let’s switch to tech, where Google is reshaping one of the most familiar interfaces on the internet: the search box. Google is redesigning it to support longer, more conversational queries and to accept more than just text—things like images, video, and files. This comes alongside a deeper blend of AI-generated summaries and traditional web links, building on Google’s AI Overviews. The company’s pitch is simple: people are asking more complex questions and want a mix of direct answers and sources. Critics, though, see real risks—less transparency about why certain information is shown, more potential harm from AI mistakes, and a major economic question: if users click fewer links, publishers and businesses that depend on Google traffic could take a hit. This is one of those changes that sounds like a small UI tweak, but could meaningfully alter how information—and money—moves online.

Google also signaled a strategic shift in the smart home. It says it’s turning Google Home into what it calls a “full-stack AI offering,” combining its Home APIs with Gemini-powered features so partners—like internet providers, security companies, and device makers—can build proactive smart-home services on top of Google’s platform. Read another way, this points to two trends: first, more of the smart-home experience may be packaged as paid subscriptions sold by partners; and second, innovation may increasingly come from third-party hardware rather than Google’s own Nest devices. For consumers, it could mean smarter automation—but also more recurring fees, and more questions about privacy and how much decision-making you want delegated to an assistant in your living room.

Finally, a major global health development: the World Health Organization has declared the Ebola outbreak in the Democratic Republic of the Congo a Public Health Emergency of International Concern, after cases were confirmed in neighboring Uganda. The outbreak involves the Bundibugyo strain, which can be highly lethal and may not be covered by vaccines designed for the Zaire strain. Health authorities stress the basics: Ebola spreads through direct contact with infected body fluids, often through caregiving or unsafe burials—not through the air—and people are not contagious before symptoms begin. The emergency declaration matters because it unlocks funding and coordination and underscores a key gap: the world still needs broader vaccines and treatments that cover multiple Ebola strains, especially in regions where cross-border movement can outpace early detection.