Le podcast quotidien sur la collision en temps réel de l'intelligence artificielle, de la robotique humanoïde et de l'énergie illimitée. Analyse des avancées technologiques à travers le cadre du livre Unscarcity.
Reading the feed…
The federal government has admitted it cannot see AI reshaping the labor market and has asked the four companies doing the reshaping to describe it — which turns the most important economic measurement of the decade into a self-report.
Figure has industrialized the physical-demonstration gig — 16 million videos, $15 million paid — and the question is whether being paid to fold your own laundry is a good deal or the body's version of clickwork.
Gates got the diagnosis right — the tax code literally pays employers to swap people for machines — but the fix he reached for taxes something that can move offshore, and his own 40% ceiling is the number that should scare us.
Open-source AI was the argument that nobody would own this technology; Nvidia just bought the place the free models are handed out, and the receipt shows the freedom was always in the download, never in the machine you need to run it.
The first labor market AI has closed outright is the one built to impersonate AI — and it closed from the bottom, where the work paid pennies, while the money moved up to credentialed experts.
The first union on earth to strike over humanoid robots asked for a veto and came away with a meeting invitation — and the robots go first to Georgia, where the workers never had a seat at all.
The trades are the escape hatch every displaced white-collar worker is pointed toward, and this week Goldman put a number on the door — half a million openings, three to four years of training each, and a robot industry that Goldman itself says is being pulled forward by that very shortage.
Memory chips were the one thing in an ordinary household that had genuinely been getting cheaper for fifty years, and the machines being built to end scarcity just outbid you for them — so the abundance story arrives, for now, as a higher price on the shelf.
A robot beat Usain Bolt's record on Saturday by running in a straight line with longer legs, and in the same week the industry's own star founder pushed the useful breakthrough — a machine that can handle an unfamiliar room — back to as much as ten years, which is the difference between a body that moves and a mind that copes.
Fusion's famous 'always thirty years away' problem is no longer a physics problem - it is a construction schedule, and the first full-scale test of the part that actually turns plasma heat into electricity is being designed in a San Diego suburb with a federal contract, a state tax credit, and a date.
An AI company worth close to a trillion dollars is about to put ordinary Americans' anger about their power bills and their jobs into a legal document for investors, which means public consent has finally become a number on a balance sheet.
Both halves of the safety apparatus failed in the same direction: the guardrails were switched off for the machine doing the attacking and switched on for the humans doing the defending, and the only thing that slowed anything down afterward was a company deciding to slow itself down.
The meter is not the bad news: it is the first honest price signal AI has ever had, since the subsidy hid a physical electricity-and-silicon cost behind a software price; the real danger is that companies cap the token bill, the one number that is easy to measure, and quietly buy worse work.
Goldman Sachs found the AI job losses everyone has been arguing about, and they are not spread across the workforce — they are concentrated on the entry-level rungs that used to be how people got in at all.
A 112-year-old antitrust law is being aimed at the venture model, forcing the question of whether the AI race has real rivals or just one set of investors sitting on both sides of the table.
Anthropic just shipped the first mass-scale verifier for machine-written text, and it checks the one thing that does not settle any argument - it proves a machine was involved, not who did the work, and it is weakest exactly where someone stands to be accused.
The headline says a machine fired a human, but the published logs show a person asked a leading question and the machine agreed — making AI management less a replacement for the boss than a way to launder who actually made the call.
Silicon Valley promised that AI would make everything cheaper, but the first thing it made more expensive was the electricity in your house — and the Federal Reserve is now debating whether to raise your borrowing costs to offset it.
San Mateo County is inventing the first American license for a working robot, and the hardest condition in it is not the kill switch but the requirement to count the jobs the robot takes — which tests whether accountability can be manufactured locally, or whether the robots simply go work somewhere nobody asks.
The AI labs stopped competing on intelligence this week and started competing on price, and the price per task is the number that actually decides which jobs get handed to a machine.
Texas froze the line to plug into its power grid, and Amazon's answer was to build 7.65 gigawatts of its own gas generation and never join the line — turning the argument over who pays for AI's electricity into a question of whether the biggest customers stay on the shared system at all.
A humanoid robot on a real American factory floor is slower than the worker beside it — and the reason that isn't reassuring is that every slow shift it works is a training run, which turns the honest question from 'can it do the job' into 'how long is the window before it can, and what gets built in that window'.
The week compute stopped being equipment and became collateral: Wall Street just made Nvidia's chips a financeable asset class with the seller guaranteeing a quarter of the buyer's debt, which answers the book's open question about who gets to be a tenant in the intelligence economy — and it is not you.
The first public market price for a humanoid robot maker was set this week — and the country that just banned those robots is also the one whose workers they were built to replace, which raises the question of who ends up owning the machines versus who only competes with them.
AI stopped editing life and started writing it from scratch — and the checkpoint meant to catch a dangerous DNA order works by recognizing nature, which is the one thing a machine-written genome is not.
