Welcome to December 30, 2025
The service economy is facing an AI margin call. Meta has effectively acquired Manus for reportedly over $2 billion, apparently valuing the startup’s ability to automate complex remote work at roughly $1 billion per percentage point of the Remote Labor Index (RLI). The RLI, a Scale AI benchmark comprising real-world, economically valuable projects across multiple sectors, saw Manus hit a SOTA 2.5%, edging out Groq 4 (2.1%) and GPT-5 (1.7%). With the market apparently pricing total remote labor automation at an outrageously low $80 billion, on margin, technological deflation is in the air.
The architecture of cognition is finding new gears. Stanford researchers achieved a breakthrough in continual learning via test-time training, allowing models to learn from next-token prediction with constant latency regardless of context length. Optimization is fractal. The NanoGPT speedrun record dropped, yet again, to 115.1 seconds. New Asian frontier labs are coming online every day. SK Telecom launched South Korea’s first 500B model, “A.X K1,” establishing a “sovereign AI foundation.” Meanwhile, corporate adoption is becoming mandatory. Notion is building custom employee agents, and Microsoft is literally paying customers to train staff on Copilot.
The gravitational pull of AI infrastructure is bending reality. Taiwanese funeral service company Tien Pin has reportedly pivoted entirely to AI chip cooling to solve thermal loops for advanced servers. The silicon fabric is densifying. TSMC has “quietly” begun volume production of N2 gate-all-around chips, claiming 15% speed gains, while Samsung accelerates its 2-nm output in Texas, raising targets to 50,000 wafers. Nvidia is cementing its hegemony, purchasing a $5 billion stake in Intel as a lifeline. Energy markets are adapting. The DOE is empowering FERC to fast-track data center power connections, while Denmark energized Northern Europe’s largest solar battery park at 200 MWh.
We are hacking the biological kernel. Sam Altman’s Retro Bio is dosing healthy humans with AI-discovered RTR242 to clear Alzheimer’s waste by rejuvenating cells. Weizmann researchers found “DARE” cells that survive heavy irradiation, identifying a caspase-based “phoenix switch” for wound repair. Insilico Medicine IPO’d in Hong Kong to double its pipeline. The FDA is contracting directly with VCs to bypass bureaucracy, while Fractyl Health seeks approval for a device that mimics GLP-1 drugs by rewiring the duodenum via endoscopy. Even lower back pain is being solved; Vertanical’s cannabis extract beat opioids in Phase 3 trials. Amidst the upgrades, US life expectancy hit a record high.
We are also finally mapping the hardware of the mind. KAIST researchers identified the prefrontal cortex mechanism for separating goals from uncertainty, extracting a blueprint for more flexible AI. Harvard’s new “SmartEM” technique has accelerated connectome imaging by 7-fold. The physics of information is burrowing into the economy. While MIT develops quantum-secure multi-party deep learning using light, Element Six is pivoting from drill bits to quantum diamonds.
The Space Age is back with a vengeance. JAXA and Toyota are building a pressurized “space station on wheels” for the Moon, securing seats for Japanese astronauts. Closer to home, Jetson held the world’s first flying car race, Amazon flying drones are delivering in Detroit, and the RAI Institute built a robotic mountain bike that combines wheels and legs.
Some human operators are getting squeezed. Resumes now have a 0.4% chance of success as AI spam floods recruiters, while the accreditation layer collapses. The world’s largest accounting body, the ACCA, has ended online exams because AI makes cheating undetectable. Yet, the loop is creating new high-value nodes. Expert annotators are now earning $90/hour to feed the machine via platforms like Mercor.
Money is becoming programmable. China’s central bank will pay interest on digital yuan, transitioning it to a deposit currency, while US AI startups raised a record $150 billion this year. Meanwhile, in the Amazon, stingless bees were granted legal rights.
Don’t blink, the Singularity moves pretty fast.



The Meta/Manus valuation framework is kinda fascinating when you look at it through the RLI lens. Pricing total remote labor automation at $80B feels absurdly low, especially considering how fast that 2.5% SOTA numer could compound with iterative improvements. I worked adjacent to an outsourcing firm a few years back and the profit margins on routine cognitive work were way higher than physical logistics, so the deflationary pressure here is gonna be pretty asymmetric. The fact that Samsung and TSMC are both racing to push 2nm production while Nvidia consolidates chip supply with the Intel stake suggests peopl are betting infrastructure bottlenecks matter more than model improvements at this point.
stingless bees were granted legal rights.
SOMETHING THE PALESTINIANS ARE BEING DENIED.