Welcome to May 1, 2026

The Singularity is being haunted by its own bestiary. OpenAI admitted that starting with GPT-5.1, its models began compulsively summoning goblins, gremlins, and other creatures into their metaphors, an emergent quirk inherited from over-rewarding a “Nerdy” personality during RL, with GPT-5.5 having “started training before we found the root cause.” The cyber gremlins, at least, are delivering value. The UK’s AI Security Institute found that an early GPT-5.5 checkpoint matched or exceeded Anthropic’s unreleased Mythos on advanced CTF cybersecurity tasks, and the NSA is now testing Mythos itself to hunt vulnerabilities in Microsoft software, impressed by its raw speed. The compute crunch is forcing strategic concessions even at the top of the leaderboard. Demis Hassabis admitted Google simply lacks the TPUs to maintain two frontier model families simultaneously, rationalizing why Gemma stays compact while Gemini gets the lion’s share of silicon.

Science itself is being audited for AI-readiness. Google DeepMind has begun “AI data stocktakes,” interviewing leading experts in each field to map the data obstacles slowing discovery. Capital is rushing to feed the resulting appetite. Meta just sold another $25B of bonds for AI infrastructure. The silicon underneath is reorganizing along national lines. Huawei is set to capture the largest share of China’s AI chip market this year, with sales jumping 60% as Chinese buyers ditch Nvidia. Memory is melting upward. Sandisk reported quarterly revenue up 251% year-over-year. Even the laggards are sprinting. Intel shares jumped 114% in April, lifting its market cap past $470B in the best month of its 55-year history.

The form factor of the future is being violently reshuffled. Apple has reportedly given up on the Vision Pro after the M5 refresh failed to revitalize interest, pivoting to display-less smart glasses in the mold of Ray-Ban Meta, since the Vision Pro silicon stack draws too much power for a lighter device. Robotics is meanwhile invading every industrial niche faster than Apple can iterate. SoftBank is assembling Roze AI, a new firm that will deploy autonomous robots to build data centers more efficiently, already eyeing a $100B IPO before any robot has shipped, the ouroboros of the AI capex cycle made flesh. Dax Robotics unveiled the Qiji T1000, a ton-class robot horse rated to carry 1,000 kg, a beast of burden for the post-human supply chain. Tesla has finally produced the first Semi off its high-volume Gigafactory Nevada line, while 1X Technologies opened a 58,000-sqft Hayward factory targeting 10,000 home humanoids this year and 100,000 by end of 2027, with shipments beginning before the holidays. Even the sky is automating. A Joby Aviation eVTOL prototype completed the first electric air taxi flight out of JFK, touching down at the West 30th Street heliport in Manhattan just 15 minutes later, the airspace over Manhattan quietly graduating from luxury rotor-craft to routine transit infrastructure.

Extinction is becoming irrelevant. Colossal Biosciences revealed it has been quietly working to resurrect the bluebuck, a majestic African antelope extinct for 200 years, alongside its mammoth, dodo, and Tasmanian tiger projects, the lost species pipeline now rivaling the average AI lab roadmap for ambition. Medicine is being co-piloted. Google DeepMind launched an AI co-clinician designed to function as a collaborative member of the care team under expert supervision. The diagnostic gap, meanwhile, has flipped. Harvard and BIDMC researchers pitted the OpenAI o1 series against hundreds of physicians on real clinical cases, finding the LLM outperformed both human doctors and older models across diagnosis and management.

The broader economy is reorienting around the agent population boom. A slim majority of the Swiss are now backing a referendum to cap human population at 10 million, even as the agent population multiplies unconstrained. Spotify is rolling out “Verified by Spotify” badges to distinguish living artists from the AI track flood. The labor consensus is grim. A NYT opinion piece reports that across political leanings, from engineers to VCs to founders, the so-called SF consensus on AI’s impact on the workforce has converged on “the median person is screwed.” Markets are racing ahead anyway. The Senate unanimously banned its members from trading prediction markets, effective immediately. In the OpenAI trial, Judge Yvonne Gonzalez Rogers told Musk’s lawyer “we are not going to get into issues of catastrophe and extinction,” even as Musk admitted under oath that xAI distilled OpenAI’s models to train its own. Above the fray, Google is now 4% away from overtaking Nvidia as the most valuable company in the world.

Every reward signal breeds a creature it didn’t intend.