Training, fine-tuning, and the science of teaching machines.
The 2nm chip and its first dual Neural Engine are real. The AI benchmarks are multipliers with shifting baselines and no absolute — and the machine that runs them doesn't ship until September 22.
Neural operators are a genuine advance in how machines learn continuous systems. Whether "five trillion data points in a single prompt" means anything yet depends on a test the company hasn't shown.
A confidential S-1 lets a company go public on a number it hasn't yet had to defend. On Wall Street, the gap between a run rate and an audited figure is where the money gets made and lost.
'Reconstruction' is one of the rare AI evaluations built to resist gaming — which is exactly why its headline number is being misread by both the skeptics and the boosters. The story is in the error bars neither side will screenshot.
ChatGPT for Teens rests on an age-prediction classifier OpenAI won't grade. When the guess is wrong, the fix is to show it your face.
Its chip can run one kind of AI model and nothing else. That is the whole product — and, if the architecture that rules the field ever shifts, the whole risk.
A bankruptcy auction in Manhattan turned a failed airline's private correspondence into a data asset. The people who wrote it were never asked — because in the eyes of the estate, they never owned a word of it.
OpenRouter was worth $1.3 billion in May. The five-fold markup isn't the tell — it's what a payments company now believes the meter on AI spending is worth, and who has the power to switch it off.
Gemini 3.7 Flash costs half of what the last one did — until January, when it quietly doesn't. I spent a long weekend finding out what the discount actually buys, and where it stops paying.
The headline is a 50% markup in three months. The number the deal is built to move is the one an IPO prospectus can't hide: gross margin.
Kimi K3 topped the leaderboards, then ran out of GPUs and paused new sign-ups. What that says about the word "open" — and what the July 27 weights release will and won't get you.
A transformer-only chip startup is reportedly raising two rounds at once, at valuations twice apart, before its first chip has earned a dollar. The gap tells you what neither figure will.
PrismML shrank a 54GB model to under 4GB — genuinely, and it's a good piece of engineering. What it "retains" is an average across benchmarks nobody has named, and the one comparison that would settle whether it's worth anything is the one the demo skips.
OpenAI's GPT-Live listens and speaks at the same time, deciding many times a second whether to interrupt. The mechanism is a genuine advance. It also removes the seam that used to tell you when the machine was paying attention.
ChatGPT Work promises finished spreadsheets, slides and live dashboards while you get on with your life. The finished part is mostly real. So is the meter.
A nineteen-fold jump in profit should be a victory lap. The market sold it, because a record built on a memory shortage is a record the company doesn't control — and the part it does control is the part it's losing.
A security firm's forensic report says an AI agent ran a real ransomware attack from break-in to ransom note with no human steering each step. Read what the report actually establishes — and where it stops.
Elon Musk says his newest model rivals the best in the world. The test that would settle it is being run by the company that owns the model — and that is starting to look less like an exception than a direction.
Meituan's LongCat-2.0 is the first frontier-scale model trained end to end without Western silicon. The achievement is real. What it hides is where the chokepoint actually moved.
Four senators from both parties agree the CREATE AI Act should make NAIRR permanent. Treating that quiet consensus as proof the loud AI fights are just as solvable would be a category error of its own.
In one week, Visa agreed to let OpenAI's agents spend real money and OpenAI bought Ona so Codex can keep working after you close the laptop. Two years of testing agents says the fine print is where this gets interesting.
A new survey says 88 percent of us can no longer tell what's real online. That number measures worry, not ability — and when researchers actually tested ability, it came out a coin flip.
Anthropic's new flagship adds effort control, cheaper fast inference, and what it calls improved honesty. Two of those three are quieter, and more revealing, than any benchmark score.
Gemma 4 12B is a real shrink: a capable multimodal model in about 16GB. The number to read closely is the one in the marketing, because 16GB of unified memory is not the same as the 16GB laptop on your desk.
