Margins tell everything
Margins are the clearest confession of what a company wants, and Nvidia could cut them and still profit – so nothing but the choice itself stops the market from flooding with builders.
Open labs need a contributor endpoint
Meta now sells a “contributor” endpoint – roughly 5% of the normal API price in exchange for storing, sharing, and training on your prompts. That’s the great move, and every open-weight lab should copy it: a honest direct endpoint specifically to get datasets in exchange for discount.
Stop downloading the same model twice. Do this instead
Every small model tweak ships as another full copy of the weights, so you store and download terabytes twice for a few changed vectors. Ship the original plus the change instead – binary diff today, LoRA file tomorrow, per-request patch in the dream.
Luddite mathematicians
Mathematicians act like luddites because their vanity and pride got hurt, while many ignore the real problem – that science gets privatized – and they refuse to change and accept the new era.
For whom the bell tolls: the copycrab music industry
Shows how record labels killed and absorbed the first generation of music-generation services, then argues for models that never train on copyrighted works – reference albums supplied at inference time, open training data, and judge models built from slop and human ratings instead.
When robots run amok
Argues that runaway drones and machines will become as routine as petty crime, so cities need robot-response teams built like firefighters rather than police to combat a new nuisance.
The sane AI slow proposal
The author proposes capping any company at 10 percent of server-grade GPUs, with progressive taxes on excess, to prevent AI monopolies, preserve competition, and force algorithmic progress rather than compute brute force.
Which side would you pick?
Argues countries should align with open AI ecosystems rather than US proprietary stacks, framing the US demand to pick sides as hegemonic coercion and open source as protection against vendor lock-in.
You should protest against the datacenters
Argues protesters should oppose data centers not for taxes, water, or noise, but because closed models fence off public resources and concentrate control over essential compute infrastructure.
Cloud AI is the worst
Argues that per-use metering destroys AI’s value by making users ration their own curiosity and experimentation, and calls for fixed-cost hardware and cheap, owned compute instead of cloud pay-per-use.