Top influences of 2025
Content that changed my life this year, and which also might change yours
2025-11-09 — 2025-12-10
Wherein the year’s leading influences are enumerated from wide reading, with AI tools reported as primary aids to reading, and NeurIPS, ICLR, human reward tampering and Coasean bargaining are recorded as recurring themes.
Assumed audience:
Laypeople and also nerds like me
I read a lot this year, as I do every year, but more than ever I used AI tools to help with much of my reading. Some of it was useful in specialist ways; I don’t expect many people will care. Some of it, though, made my life different, so I’m mentioning it here. I “agree” with relatively few of these pieces entirely. The criterion for inclusion is that the piece influenced me, not necessarily that it agrees with me. That is to say, this list does constitute an endorsement of quality, but not of the content as such.
Not all of this was written this year. I don’t read the whole internet the moment it updates.
1 Conferences, symposia, workshops
ILIAD2 was a theoretical alignment unconference that totally shifted my thinking about credible approaches to AI alignment. Amongst other things, it persuaded me there might be content in Singular Learning Theory.
2 Civitech
Digital governance and civitech and other notions of what social coordination could or should be in the age of AI
I am spinning up a modest civitech project at the moment in my home, Melbourne/Naarm.
3 Oh wow it turns out that Human Computer Interaction is not only real it is existentially important
I always thought HCI was too soft and pre-paradigmatic to cohere into a real science. And, too vague to make useful predictions.
I was wrong. Turns out it’s the discipline that studies the interface between human and machine intelligence, and as such is crucial for our survival. Moreover, it turns out that although the median HCI paper might be a bit meh, the best ones are really good.
You can see what I mean by looking at some of my recent funding proposals, on human reward tampering, human-AI asymmetries …
I not hold that actually-existing only makes sense as a coupled human-AI system, and we need to understand what that system does.
Relatedly…
4 Societal Epistemic Health
How do we maintain a healthy infosphere in the sense of having good beliefs and avoiding bad beliefs, in society at scale, with ubiquitous deep fakes and LLMs that can generate plausible-sounding bullshit?
Some readings that helped me think about this:
5 Stochastic parrotology
Ornithology of aleatoric psittaciformes. Chineseboxology.
These are abstruse ways I might describe the questions of predictive pretraining versus agency. Can pretraining lead to understanding? Can it lead to agency? When?
6 Political Economy of Actually-existing AI
The hype/delivery landscape is fractally complex, and anyone’s take that reads like “it will obviously go like X” isn’t credible to me — they are merely feeding another hot take into the hot take furnace.
Some types of hype are predictably nonsense. Never before in human history has it been so painfully clear that whichever institution it is that issues qualifications for “thought leadership” really needs to work on its syllabus.
That said, certain analyses I think are worth paying more attention to for their ability to help us discern the possibilities. Here are some pieces that attempt to bring at least some vague knowledge of technical and or socio-economic systems to bear on the political economy of AI.
- I Will Fucking Piledrive You If You Mention AI Again (on detecting when you are being sold hype that cannot deliver)
- The Electric Slide — by Packy McCormick and Sam D’Amico (On backing out the bits-vs-atoms strategies pursued by the two global superpowers)
- The Political Economy of AI: A Syllabus - by Henry Farrell. I picked this at random from a few interesting articles in the political economy rubric.
- Inside the Tony Blair Institute (Tech industry influence and/or state capture in middle powers, a case study)
- Economics of Transformative AI Workshop, Fall 2025
7 Thomas Urquhart
Guys, this dissolute, self-sabotaging, anarchic genius has all his works online and I missed it. Did you know he wrote a proposal, Logopandecteision, for the ultimate language in which all words described themselves by their sounds, but did not finish it because most of the book was him complaining about his creditors? Did you know he invented a language, Trissotetras, whose utterances are trigonometric proofs? Quixotic mathematical-linguistic genius whose time has finally come.
- Maligned for mathematics: Sir Thomas Urquhart and his Trissotetras: Annals of Science: Vol 76, No 2
- The Trissotetras: Or, A Most Exquisite Table For Resolving all manner of Triangles, whether Plaine or Sphericall, Rectangular or Obliquangular, with greater facility, then ever hitherto hath been practised. Most necessary for all such as would attaine to the exact knowledge of Fortification, Dyaling, Navigation, Surveying, Architecture, the Art of Shadowing, taking of Heights, and Distances, the use of both the Globes, Perspective, the skill of making the Maps, the Theory of the Planets, the calculating of their motions, and of all other Astronomicall computations whatsoever. Now lately invented, and perfected, explained, commented on, and with all possible brevity, and perspicuity, in the hiddest, and most re-searched mysteries, from the very first grounds of the science it selfe, proved, and convincingly demonstrated
- Urquhart’s Inflationary Universe | SpringerLink
- “Verbs, mongrels, participles and hybrids”: Sir Thomas Urquhart of Cromarty’s Universal Language in: “Joyous Sweit Imaginatioun”