Goodness me. Fascinating.
Hospitals using AI are seeing better healthcare outcomes.
In 2023, health systems with high and low AI use had the same death rate: ~1,600 per 100,000 patients.
By 2026, the systems using AI the most dropped to ~1,000 deaths per 100,000.
That's a decline of nearly 40% in just 3 years.
Meanwhile, hospitals with lower AI adoption have seen significantly smaller improvements.
The gap in patient outcomes is widening as AI adoption accelerates.
AI is rapidly transforming healthcare.
Kyle Saunders retweeted
So it happened. This is the first time I've given Pangram a text entirely written by me, about my own life, and it said it was AI-written (58% of it!). I'm sorry, @pangram, but this false positive should be enough to stop this from ruining people's literary careers on probability
Kyle Saunders retweeted
"Are we getting enough Republicans?" is the wrong question, historically speaking.
Polls have always missed some segments of the population -- notably young and non-college. But it used to matter less because those people either didn't vote or were split between parties.
Problem in the last ~10 yrs: Education polarization transformed a non-problem into a major problem that favors one side.
We can weight up non-college to the correct proportion of the sample, but if the nonrespondents systematically differ from respondents (they do), we could be magnifying the impact of nonresponse.
This is why I worry way more about Republican underestimates than Democratic ones -- it's not actually about the party. It's about populations we've never been good at capturing and which direction they lean.
I'm not sure what it would take to de-entrench the presumption that polls systematically underrate conservative parties or candidates. It's not true historically, but it's a cheapie take that will never really get you criticized.
web.archive.org/web/20170423…
Tuesday, Oct. 27, 11 a.m.: I’m giving the keynote at CSU’s CLA Democracy Summit, “The Political Crossroads of Artificial Intelligence.” Info link below.
Americans aren’t anti-AI per se. They use it more every quarter, and big bipartisan majorities still want it slowed down and checked.
They’re against AI that nobody answers for.
This fall, most of the major AI decisions got made over a White House lunch, in executive orders and in lawsuits, while some of the most accountable ones are happening at city council meetings a couple of miles from campus.
A week before an election, the question is who gets a day in deciding what happens next.
libarts.colostate.edu/resear…
If you're going to be in Fort Collins on Tuesday, and are not doing anything at around 11am, and are in the mood to go down a nerdy rabbit hole with me, well, you might as well come see me give the keynote to Colorado State University's Democracy Summit. More information here:
libarts.colostate.edu/resear…
Kyle Saunders retweeted
An eye-raising new lawsuit filed this evening: two former and one current NCAA Division III athletes filed suit against the NCAA over Division III rules barring athletic scholarship, claiming that the restriction amounts to illegal price fixing under antitrust law.
Kyle Saunders retweeted
NEWS: A class action antitrust lawsuit has been filed against the @NCAA challenging its ban against athletic scholarships at the @NCAADIII level. Lawsuit comes 13 years after a previous legal challenge over D-III financial aid fell short. For @Sportico👇
sportico.com/leagues/college…
and that was Opus 4.8...(we're on Opus 5.5 now.)
Kyle Saunders retweeted
Massive. New Harvard study finds that AI-written paper reviews now beat the best human peer reviews on every measure the authors tested.
Peer review is slow and often misses errors. Earlier studies concluded AI should only assist human reviewers, not replace them.
They used 134 reviews from rapidPeer, a Harvard service that pays expert reviewers well. Claude Opus 4.8 reviewed the same 1124 papers, and GPT-5.5 plus 3 former journal editors judged the pairs blind.
GPT-5.5 picked the AI review in all 132 decided pairs, and the editors picked it 87% of the time. Only 5.5% of the AI's claims about the paper were wrong, versus 15.9% for the human reviews.
Teams that depend on expert critique can now let AI do the first full review pass, and keep a human for the final decision.
Kyle Saunders retweeted
Very interesting. I've been waiting for this kind of more detailed analysis of Claudelish and the like.
Kyle Saunders retweeted
People are getting smarter
Kyle Saunders retweeted
We don't have a lot of evidence, but the bulk of the academic papers modelling the dynamic seem to suggest AI initially helps attackers more than defenders because of the low marginal cost of attacks in the short term.
Kyle Saunders retweeted
The intelligent @profgoose introduced us non political people to a concept called rider. I talked to some folks inside athletics yesterday about the PCSA and they feel the only way to get it passed is to attach it to another bill - the rider.
Kyle Saunders retweeted
New paper: "AI Article Reviews Dominate Human Peer Reviews at Their Best: Evidence and Policy Choices" See: GaryKing.org/peerless. with Steven Worthington. Comments welcome!
1/ Most university AI task forces still leave the student AI rule to each instructor, administrations keep building committees, students barely have seats, and a small few now say try a detector. The inventory is up to 106 reports at 81 schools. kylesaunders.com/ai-taskforc…
4/ Oregon's AI Council says the university discouraged detectors "up to now" and should be ready to pilot the newest tools. That makes three reports that recommend trying one, against 35 that oppose them or warn about them.
5/ A second crawl of all 372 schools re-found reports at 39 of the 55 schools where I knew of one, against 44 in September, with essentially the same code. One crawl is a sample, and the site says so. Missing your school's report? Send the link. Write-up: kylesaunders.substack.com/p/…
