iiinova — where your digital self keeps evolving

‮Jeff Eaton @eaton.fyi · 02/10/2026

The hardest part of explaining politically active reactionary religious fundamentalism to people is convincing them that “hypocrites” is just an irrelevant label

020

‮Jeff Eaton @eaton.fyi · 01/10/2026

This is one of the reasons apocalyptic language and belief are red flags; “apocalypse” in fundamentalist discourse isn’t “the end of the world,” it’s “God wipes out everyone else and we are proven right and good.”

‮Jeff Eaton @eaton.fyi · 29/09/2026

LLMs just drift in the direction you go, because they do not *have* communicative intent.

‮Jeff Eaton @eaton.fyi · 29/09/2026

One of the things i’ve been thinking about is that LLMs, unless very carefully steered otherwise, fail to perform a key function of conversation: “repair.” When you notice someone is misunderstanding you or is unconvinced and you adjust to communicate more effectively.

142

Oh yeah, I jumped straight to the one that most people are willing to argue about rather than hedge, but it goes down to that eventually

Not in some hand wavey definitional sense, but in the real “just as legitimately entitled to chart our shared course as I am” sense

Real coexistence is tough because it requires a degree of acceptance that The Other is equal.

Yeah. And the degree to which the particular cluster of people being discussed envision that as “this compound” vs “this community” vs “this nation” vs “earth” is important. Even listening to a dyed in the wool white nationalist like Doug Wilson will jump between those when it serves him

Both, and there’s a permeable barrier between the ideas rather than a hard line. Some insist the United States *is* that separate place.

I say that not because I’m a doomer, but because I spent a good chunk of my life in a group that fit that description. If part of your society views coexistence and equity as a failure, not a victory, that is a first-order alignment problem to iron out, not an implementation detail.

2127

“How do we live together“ is not the same question as “How do we live together with people who don’t want to live together with other people”

1123

I don’t think people are starved for that. I think humanity had been thrashing out the details for a while, with varying degrees of success, and our current moment is what it looks like when wreckers don’t *want* to live together ethically and morally with anyone else.

1113

This comment here absolutely defines the dangerous part of assistive AI. Figuring things out *is* a rush. When you’re learning, that’s the feedback loop that keeps you going, and eventually gets you to the place where you can make and own the more nuanced and complicated decisions.

040

Vernor Vinge’s narrative aside in A Deepness In The Sky — thousands of years of software cruft, impenetrable and unpredictable but mediated by AI assistants — is feeling a little too close. Interesting that Pham, the book’s center of gravity, becomes the master of an army simply by digging deeper.

040

Kevin Riggle @kevinr.free-dissociation.com · 25/09/2026

I think we can all agree that if Community were still going, any problems would be overshadowed by how amazingly perfect the “Dean Pelton Gets AI Psychosis” episode was

2131

Reposted by ‮Jeff Eaton

As a cybersecurity consultant, people are constantly asking me how they can have their cake and eat it too, and the answer is: I'm sorry, but sometimes there are hard trade-offs, and here you simply can't.

3353

‮Jeff Eaton @eaton.fyi · 24/09/2026

I mean if something like that happened you would see it in other statistics like long term health impact for those affected ¯\_(ツ)_/¯

0620

‮Jeff Eaton @eaton.fyi · 21/09/2026

Next year snowfall will be old enough to drive, man

070

‮Jeff Eaton @eaton.fyi · 21/09/2026

Hypercalvinism with GPUs and a trilby.

000

‮Jeff Eaton @eaton.fyi · 19/09/2026

This is a terrible time for systems thinkers, in part because nobody wants to listen to systems thinkers about stuff they’ve been saying for years, so now systems thinkers have to watch deeply incurious people reinvent systems thinking from first principles

39315

‮Jeff Eaton @eaton.fyi · 18/09/2026

Imagine the entire automobile safety universe was dominated by Cars Truthers. Like, people who believe the movie Cars is a documentary. Everything they write about crash testing, engineering standards, etc has this weird tinge and you can’t put your finger on it until… Oh. Oh no.

061

Reposted by ‮Jeff Eaton

Jerry Chen @jcsalterego.bsky.social · 18/09/2026

when the moon hits your eye like a big pizza pie, oh god that's 90 billion megatons of TNT all civilization is wiped out earth plunges into fiery inferno and then utter darkness

4338126

‮Jeff Eaton @eaton.fyi · 17/09/2026

Oh look, the latest issue of the preeminent semi regular outsider zine on the topic of the hypertext markup language

Issue 2 of “Multipage Version,” a zine about html

0163

‮Jeff Eaton @eaton.fyi · 17/09/2026

that’s not very bayesian of you

‮Jeff Eaton @eaton.fyi · 17/09/2026

“[T]here is another way to make $300,000,000. Start with $30,000, and using a quantum random number generator, gamble it on the forex markets at a 10,000:1 ratio” You know, It’s been a long time since I read it, and I had completely forgotten how mind numbingly stupid that discourse sinkhole is.

181

‮Jeff Eaton @eaton.fyi · 16/09/2026

what if the texas sharpshooter fallacy, but AUTOMATED

‮Jeff Eaton @eaton.fyi · 16/09/2026

novelty barbarian/bard multiclass here he comes

fluffy Siberian cat named Felix, surrounded by a pile of novelty dice

0130

‮Jeff Eaton @eaton.fyi · 15/09/2026

Just imagine how this would’ve gone if yud was a narnia guy

220

‮Jeff Eaton @eaton.fyi · 15/09/2026

the most exhausted imaginable “yes, this”

071

‮Jeff Eaton @eaton.fyi · 13/09/2026

This. If validation is roughly as expensive as generation, the ROI doesn’t materialize

071

‮Jeff Eaton @eaton.fyi · 12/09/2026

I hear that. I feel like the aughts and teens either forced a serious hard think for most of us or just turned us into gleeful sociopaths

‮Jeff Eaton @eaton.fyi · 11/09/2026

one fun part of using your own public content creations over the years as a demo data source for an llm processing pipeline is being reminded of just how many of your public taglines and profile blurbs are prompt injection attacks

250

‮Jeff Eaton @eaton.fyi · 11/09/2026

p sure @theletterf.bsky.social has taken that route or something adjacent using LLMs to transform styleguides into vale schemas for later deterministic processing

‮Jeff Eaton @eaton.fyi · 11/09/2026

probably one of the first ai things I went on the record with a couple years ago, when people started talking about how AI would kill "design systems" because it'd free us from the need for constraining ~systems~ when building, and, uhhhhhh... autogram.is/discussing/p...

autogram.is

Pattern Wise, System Foolish — Autogram

Design systems are not pattern libraries, and abandoning them for "experimental" AI won't free us from constraints.

111

That’s exactly the trick. And once you start speaking a specific formal vocabulary to improve the parsability, you’ve got humans doing the hard part again, COINCIDENTALLY

170

Yep. And my corollary: If validating quality or correctness in your domain is roughly as hard as producing it, you will not see ROI with generative tools

Lotta work with AI stuff for the job of late. And by that I mean “being pleasantly impressed with demo level output and unshocked at the limits of scaling it to real-world levels of contextual ambiguity, then slogging through the work of formalizing the connective tissue that human teams intuit”

1110

Which brings us back to “the fields that have demonstrated measurable outcomes from generative ai tools are ones that had pre-existing existing linters”

I’m sure we could just point claude at the existing research and say “make that a skill,” that’s how that works, right?

100

Even worse: “Rich guy says thing” doesn’t have to imply that for the diffusion of noise to effect the information environment. Again tapping the Jaques Ellul sign and sighing and reminding myself that propaganda doesn’t *convince* as much as it *normalizes and crowds out*.

Seems to be some prior art, natch www.arg.tech/index.php/pu...

arg.tech

Publications

And as we have seen, “listen to the experts” collapses in a world where competing ideas bring their own armies of experts and monotruth is deliberately undermined.