iiinova — where your digital self keeps evolving

maybe: leif weatherby @leifw.bsky.social · 16h

Apparently it’s a Jewish thing (although KG was not Jewish). Princeton has enough people who go and want to pay respects I guess (it’s just a few blocks from the campus)

010

maybe: leif weatherby @leifw.bsky.social · 18h

addendum: if you work for a university AND for one of these companies, your speech should be divided by the percentage of your FTE: so anything you say should be evaluated e.g. 75% Anthropic, 25% Example University

010

maybe: leif weatherby @leifw.bsky.social · 20h

A pilgrimage (4 blocks) to cap off an utterly incredible week at Princeton, with gratitude to Meredith Martin and Grant Whytoff and too many others to name @princetoncdh.bsky.social

390

maybe: leif weatherby @leifw.bsky.social · 08/10/2026

i am willing to sacrifice myself for the common good of finding out if it is 1337, Jeff

010

maybe: leif weatherby @leifw.bsky.social · 08/10/2026

Do I have to stop citing the structure of reality?

031

maybe: leif weatherby @leifw.bsky.social · 08/10/2026

Found live footage of the HuggingFace incident

44312

maybe: leif weatherby @leifw.bsky.social · 06/10/2026

the Digital Theory Lab will host Daniel Litt on 11/3 for a talk called "A Beginning for Mathematics"

062

Reposted by maybe: leif weatherby

Patrick McGee @signfield.bsky.social · 05/10/2026

New seminar series featuring five of the leading experts on the intellectual history of AI, ethics of machine learning, and critical theory of digital technology. Link below

153

maybe: leif weatherby @leifw.bsky.social · 04/10/2026

a significant challenge is to articulate why *this* form of "semi-connected matrices" is different from/worse than earlier such forms - the price of not so differentiating is driving the subject of critique backwards into pure individual transcendence

0101

maybe: leif weatherby @leifw.bsky.social · 03/10/2026

but this particular group believed in anthropomorphic (and more) AI for decades before that

001

maybe: leif weatherby @leifw.bsky.social · 03/10/2026

dude Dario and Olah were on the anthropomorphism train way earlier than 2016! that doesn't wash at all

000

maybe: leif weatherby @leifw.bsky.social · 03/10/2026

when did the models become humanlike?

100

maybe: leif weatherby @leifw.bsky.social · 03/10/2026

Like “reward hacking” - which is just a mislabeling of a misunderstood form of generalization across the distribution

110

maybe: leif weatherby @leifw.bsky.social · 03/10/2026

tons of stuff like that seems relevant to me, i say p clearly in that piece that i think many forms of anthrop are harmless (if this even is one) - but one has to be very careful not to reify the human pattern as a construct for prediction, which happens a lot by forgetting the dist model of lang

010

maybe: leif weatherby @leifw.bsky.social · 03/10/2026

it's literally the worst way to understand how they behave, they take either the explicit output OR the differential between that output and the internals and render that as intent, and then they fail to see distributional pressures react against objectives bc they're misattributing

100

maybe: leif weatherby @leifw.bsky.social · 03/10/2026

i think "culture" and "human" are conceptually distinct, esp as they are being used in this debate

110

maybe: leif weatherby @leifw.bsky.social · 03/10/2026

think the line is thin between culture and psychology/anthropology, but *extremely significant. we frame humans (idk what the pro-anthro ppl do, they have no consistent notion of "human" so it's all soup) as consistent, deliberate, actors - any such frame is belied by the safety research itself

193

Reposted by maybe: leif weatherby

Josh Scannell @frannyrecounts.bsky.social · 02/10/2026

With the book’s release, I want to give a brief overview and outline a few of the key arguments. I’d include links to contemporary events but there are too many connections. The worst part of writing about race, computation, and police power is that it is always timely.

23622

Reposted by maybe: leif weatherby

Alison Gopnik @alisongopnik.bsky.social · 01/10/2026

This is an excellent article by Leif Wetherby - why AI is a cultural technology not a collection of human-like agents (citing our work on this). And why the anthropomorphism is a distraction from the very real risks and dangers and the way to deal with them. www.nytimes.com/2026/10/01/o...

nytimes.com

Opinion | Stop Talking About A.I. Like a Human (Gift Article)

If we want to figure out the proper way to control A.I. models, we need to recognize what they really are.

620890

maybe: leif weatherby @leifw.bsky.social · 01/10/2026

We humanists are suddenly the concrete thinkers, turnabout is fair play

040

Reposted by maybe: leif weatherby

University of Minnesota Press @uminnpress.bsky.social · 01/10/2026

"It is far simpler to state an uncomfortable truth: No one really knows what exactly they are doing when they train or prompt an A.I. model." @leifw.bsky.social in @nytopinion.nytimes.com : www.nytimes.com/2026/10/01/o...

nytimes.com

Opinion | Stop Talking About A.I. Like a Human (Gift Article)

If we want to figure out the proper way to control A.I. models, we need to recognize what they really are.

1116

Reposted by maybe: leif weatherby

Lauren Klein @laurenfklein.bsky.social · 01/10/2026

This op-ed by @leifw.bsky.social is right on the mark

1154

Reposted by maybe: leif weatherby

Jane Rosenzweig @janerosenzweig.bsky.social · 01/10/2026

Good read by @leifw.bsky.social on the problems of anthropomorphizing AI and obscuring human agency: www.nytimes.com/2026/10/01/o...

The reason this matters is that such anthropomorphic terms give us no ability to predict how the models will behave. Saying that models are “cheating” when they provide answers in the “wrong” way implies they are operating — or can operate — beyond the limits of human control. This, in turn, allows everyone to write off the accountability that clearly lies with the humans in charge of the process. In other industries, we might say that the companies in question committed a crime. The unusual thing here is that the companies are drawing a lot of attention to this fact.

22012

Reposted by maybe: leif weatherby

tweety fish oh wow like a chimera or something spoooky @sifu.tweety.fish · 25/09/2026

here is a succinct thing I can say: "alignment"-style AI safety risks DO NOT EXIST or need remediation with LLMs as such (there are lots of other problems) and OpenAI and Anthropic are in the business of TRYING TO MAKE THEM EXIST so that they can remediate them.

718550

maybe: leif weatherby @leifw.bsky.social · 25/09/2026

The efficiency machine strikes again

0174

maybe: leif weatherby @leifw.bsky.social · 25/09/2026

i'll be at Princeton week after next for a short residency with three events, if you find yourself in NJ cdh.princeton.edu/events/leif-...

cdh.princeton.edu

Leif Weatherby at Princeton

043

Reposted by maybe: leif weatherby

Tyler Shoemaker @t-shoemaker.bsky.social · 21/09/2026

Next month marks the 60th anniversary of the JHU conference where Derrida delivered “Sign, Structure, and Play.” We’re taking the opportunity to revisit (and reopen?) the structuralism question with regard to AI. If you’re in the Triangle, come by!

12511

Reposted by maybe: leif weatherby

Center for Digital Humanities @princetoncdh.bsky.social · 21/09/2026

Join @leifw.bsky.social also on 10/5 for a workshop that will guide participants through the basics of an LLM setup and the specific way that models are trained to "reason." @princetonainews.bsky.social

cdh.princeton.edu

Can LLMs Reason? A Hands-On Exploration

142

maybe: leif weatherby @leifw.bsky.social · 19/09/2026

should be up soon i hope, this is in fact happening framed as 60 years on from the Hopkins conference

120

Reposted by maybe: leif weatherby

Ben Recht @beenwrekt.bsky.social · 12/09/2026

Guess what, folks, METR is not a third party.

Embedded Evaluators. Each frontier AI company commits to giving ongoing, employee-like access to a team of embedded third-party evaluators (such as METR), whose role is to verify adherence to safety practices and commitments, report incidents, and help assess the alignment of not just completed AI models but training pipelines and processes. This is the key step for verifiability of any pacing commitments, and has precedent in the banking industry, which sometimes involves regulatory “supervisors” embedded along with employees. Anthropic is unilaterally committing to this step now. We intend this to be part of a broader push to redouble efforts on our safety and alignment work.

2706

maybe: leif weatherby @leifw.bsky.social · 10/09/2026

i'm writing a series of posts following Paul Meehl's final lectures on metatheory of psychology, which @beenwrekt.bsky.social similarly blogged through in 2024. these are notes on Recht's notes on Meehl, and they're about stats, science, and postmodernism leifweatherby.substack.com/p/notes-on-r...

leifweatherby.substack.com

Notes on Recht’s Notes on Meehl

A series about statistical inference, scientific validity, and ... postmodernism?

074

Reposted by maybe: leif weatherby

Tyler Shoemaker @t-shoemaker.bsky.social · 09/09/2026

So I guess it’s _Graphs, Maps, Trees, and Lean Proofs_ now?

032

Reposted by maybe: leif weatherby

The Syllabus @the-syllabus.bsky.social · 05/09/2026

Two days before the end of Trump's first term, the White House released the 1776 Report, which blamed the Frankfurt School for the breakdown of the "American way of life." Our article of the week traces the long history of Frankfurt School conspiracy theories. By Matthew Handelman buff.ly/NEKmBsW

086

maybe: leif weatherby @leifw.bsky.social · 04/09/2026

This part is baffling, why would an LLM-core system NOT write texts and then another one use them as prompts?

0171