Founder of @MetascienceObs, researcher @RadvacProject "Speak your mind, even if your voice shakes". Leave anonymous feedback admonymous.co/delton137
Boston, MA
- Tweets 22,879
- Following 3,144
- Followers 8,182
- Likes 59,898
Yeah AI is the real deal and can be creative and solve unsolved problems. Many still don’t want to acknowledge that fact.
Update on Navier-Stokes: 🦜🦾
"On a technical level, the paper is unbelievable, maybe beyond what a human could have come up with but not beyond human understanding: there are several different groups of pde experts working on understanding it who say that they more or less understand now."
"Some on these ideas are in the literature (e.g. Leibovich-Stewartson 1983) but the way they are combined is definitely insane and definitely nowhere in the literature previously.
I also don't know of anyone suggesting that convex integration would play a role in the creation of a blow-up solution like this."
1
19
1,004
The forensic metascience AI toolkit I’ve been developing @MetascienceObs most likely would have caught this! (I’ll run it later today)
Absolutely brutal: Over four years in review with Nature. Still published with a copy and paste error between experimental groups. Checked everything but the raw data I suppose.
241
As I've been saying, every new scientific paper should undergo thorough AI peer review using a frontier model, and all major errors should be corrected *before* publication. And the AI should have forensic metascience tools, like those in the AI toolkit I created @MetascienceObs.
By every measure, Opus 4.8 beats human experts as peer reviewers in a study of 124 Harvard social science papers covering a range from statistical analyses to formal models to qualitative work.
As academics, we need to figure out how to adapt the practice of science to AI, fast.
1
1
8
603
Three talks from the event I ran in Cambridge during bostonlongevityweek.com/ are livestreaming. If you want to be informed about future events, send me a DM!
Longevity Lightning Talks in Cambridge, MA: Preston Estep, Dan Elton, and Gennady Stolyarov II nitter.siddz.com/i/broadcasts/1AGRnZDdw…
4
261
Perhaps a bit late, but I spent a couple hours with Claude this morning updating everything on my computers and picking the low hanging fruit to improve security.
1
8
496
self-recommending!
Where the Industrial Revolution Came From
@jasoncrawford and @benlandautaylor cover the Industrial Revolution, the acceleration of technology, and how our increasing awareness of technological progress as a historical force has altered society.
0:00 — Intro
0:50 — The accelerating flywheel of technological progress
8:31 — Why couldn’t the Industrial Revolution happen sooner?
14:44 — The role of philosophy of progress
19:42 — Prerequisite technology
22:09 — Why no Chinese Industrial Revolution?
28:19 — What was going on in Britain?
33:22 — The role of science
39:11 — The shadow of the World Wars
50:23 — The durable echoes of older ideologies
3
269
Dan Elton 🔍 retweeted
PDF here gking.harvard.edu/files/peer…
Every paper should be thoroughly audited and reviewed by a highly advanced AI system (not Gemini Flash!!) equipped with forensic metascience tools, and all errors it finds should be fixed **before** publication.
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.
1
7
422
Really good overview of recent AI-powered cyber attacks in this newsletter: p3humansonai.substack.com/p/…
2
2
292
PSA for Boston/Cambridge area folks - on Monday @stevenwardell and I are running an AI salon in Cambridge with @JohnClarkLevin, who has been research lead for Ray Kurzweil for ten years. He's also been doing really interesting outreach work to Catholic leadership to get them involved in AI issues, including AI x-risk. John-Clarke is visiting town from Los Angeles. RSVP here: luma.com/056kdakp
3
2
254
you can also RSVP on Less Wrong! lesswrong.com/events/DAsc3HM…
97
Dan Elton 🔍 retweeted
New on our YouTube - Metascience Observatory founder Dan Elton (@moreisdifferent) explains our fetchpdf package for downloading PDFs legally. In many cases you can download XML when you can't get a PDF & then convert the XML to a PDF!
pip install fetchpdf
piped.video/watch?v=ifjcwuPc…
1
4
323
I recorded a little video this morning explaining the fetchpdf Python package I've been working on. One of the cool things I learned is many publishers offer XML (JATS) even if they don't make available a PDF (bc paywalls). With fetchpdf you can get the xml and convert to PDF!
4
1
10
515
sad to see the "AI risk fears are just regulatory capture" nonsense being promoted by Lifeboat Foundation, which is supposed to be devoted to safeguarding humanity from catastrophic and existential risks..
55
