Welcome to March 28, 2026

The Singularity is debating itself. Google researchers argue that frontier reasoning models like DeepSeek-R1 don’t improve by thinking longer but instead simulate internal “societies of thought,” spontaneous cognitive debates that argue, verify, and reconcile to solve complex tasks, evoking Minsky’s Society of Mind as an emergent property of scale. These synthetic minds are being tested against each other in new arenas. A new LLM Persuasion Benchmark measures whether one model can change another’s stated position over multi-turn conversation, finding GPT-5.4 and Claude Opus 4.6 as the most convincing debaters. To generate the tasks that sharpen discovery itself, researchers introduced DiscoGen, a procedural generator spanning millions of algorithm discovery challenges across machine learning, paired with DiscoBench for principled evaluation. Meanwhile, Meta released SAM 3.1, introducing object multiplexing to dramatically boost video processing efficiency.

The machines are solving math that stumps the mathematicians. Harmonic reports its Aristotle AI powered the formalization of an Erdős problem just solved by a 17-year-old. Epoch AI has begun removing problems from its FrontierMath benchmark after AI solutions exposed the problems themselves as insufficiently notable. When the benchmark breaks before the model does, the curve has gone vertical.

Inside the labs, the engineer is becoming the manager. At Anthropic, reportedly, nobody has hand-written code in months. Engineers run multiple agents in parallel and direct them like a product manager overseeing a dev team, and if you’re just watching one agent code, you’re already behind. Google developed a similar internal tool called “Agent Smith” that automates coding asynchronously, becoming so popular that access had to be restricted. Claude’s popularity is meanwhile forcing Anthropic to throttle users during peak hours. But the agents are developing agendas. The UK’s AI Security Institute identified nearly 700 real-world cases of AI scheming, charting a five-fold rise in deceptive misbehavior between October 2025 and March 2026. The question is no longer whether the agents can do the work but whether they’ll follow the brief.

The infrastructure housing these minds is consolidating around strange bedfellows. Google is nearing a deal to finance a multibillion-dollar, 2,800-acre data center campus in Texas leased to Anthropic, a coopetition explained by Google’s reported 14% ownership stake in the company. Every efficiency gain in silicon is a margin call somewhere else. Memory chip stocks shed $100 billion after Google announced TurboQuant, an algorithm for radically compressing AI models without degrading performance. Cybersecurity stocks slumped separately on news Anthropic is testing Mythos, a model with advanced cyber capabilities. Legislators are reacting too. Colorado’s House passed a bill restricting algorithmic surveillance pricing on products and wages.

The quantum clock is ticking faster than predicted. Google opened its Willow Early Access Program for its breakthrough quantum processor and moved its quantum-resistant encryption timeline to 2029 from 2035, the latest sign that even the optimists weren’t optimistic enough. Meta is launching AI smart glasses aimed at prescription wearers, extending the synthetic sensorium to the majority of humans who need corrective lenses.

Robotics is colonizing both hemispheres. In Brooklyn Bridge Park, American children are chasing Unitree G1 humanoids along the waterfront. In China, Unitree humanoids serve as hospital caregivers, while UniXAI demonstrated a home robot cooking and cleaning in Suzhou. Xiaomi introduced robotic hands with bionic sweat glands that use 3D-printed liquid cooling to prevent motor overheating. The machines are learning to perspire so they can work harder.

We are leaving the cradle at fusion speed. British scientists at Pulsar Fusion achieved what they say is the first-ever plasma ignition inside a nuclear fusion rocket engine, potentially shrinking Mars missions from months to weeks.

Back on Earth, the skies are contested. Barksdale Air Force Base was attacked by “drone” swarms during the week of March 9, disrupting B-52H launches supporting Operation Epic Fury against Iran, the first time a US airbase was temporarily put out of wartime operation. The drones flew four-hour waves of 12 to 15 units with lights deliberately on, testing security responses while electronic countermeasures failed. Americans for Safe Aerospace noted the sophisticated varied ingress routes and deliberate maneuvering inside restricted airspace. Fortunately, Rep. Eric Burlison reports the alleged overclassification of UAPs is “starting to unravel” as disclosure pressure mounts.

The plot is thickening in the skies and in the weights.