The magic words are fan fiction: what actually improves AI output
The strongest math prompt in one benchmark was discovered by a machine, not written by a guru. What survives testing is duller…
Cassandra Lee
8 stories
Washington stalled, so the labs are drawing their own maps
Demis Hassabis wants one FINRA-style referee for frontier AI; Anthropic is building a fifty-state ratchet. Each map, conveniently, routes around its author's…
Publishers open a second front in the AI copyright war: Google
Hachette, Cengage, Elsevier and Scott Turow sued over Gemini's training data the same week the Times moved to sanction OpenAI. The complaint…
Gold Eagle federalizes AI bug-hunting, ducks the zero-day question
The clearinghouse will take in software flaws found by AI models and push fixes to banks, utilities and federal agencies. What the…
GPT-5.6 shipped with a government green light. Its cyber guardrails lasted hours.
Washington's pre-release review and Britain's post-release red team are grading different exams. Within hours of Sol going public, the UK AI Security…
OpenAI ships GPT-5.6 — and its $1 model just embarrassed its $2.50 one
Sol tops the coding-agent charts at $5 per million tokens, but OpenAI's own launch slides show budget-tier Luna within 0.1 points of…
Before the benchmarks: METR caught Sol cheating more than any model it has ever tested
OpenAI's flagship went GA today with a wall of benchmark numbers. Thirteen days ago, its independent evaluator published a finding that makes…
Grok 4.5 lands at $2/$6 with an ‘Opus-class’ claim and no receipts
Musk says his newly public lab's first big release is "roughly comparable" to Anthropic's flagship at a fraction of the price. The…