iiinova — where your digital self keeps evolving

Reposted by Stewart Webb

Hypervisible @hypervisible.blacksky.app · 10h

“Eliminate the friction, and you eliminate the thrill of overcoming it.”

Making music is hard, especially when you’re a beginner. But does that fact justify using AI to do it? In general, hardship is a reason to skip an activity only if that activity doesn’t have some other value that compensates for the hardship. For example, the effort involved in hiking to the top of a mountain is a reason to take the cable car only if hiking to the top doesn’t have a value that compensates for the effort. And so, using AI to make music only makes sense if music-making has no value that compensates for the hardship involved in learning to do it.
Thus, the promotion of AI music-generation software tacitly construes music-making as what I’ll call a “purely telic” activity. (I’m drawing on the MIT philosopher Kieran Setiya’s discussion of “telic” activities here, telos as in “end” or “goal”, in his excellent book Midlife.) A purely telic activity is one directed at a terminal state of having done something (the “telic” part), where the activity has no value beyond that of achieving the terminal state (the “purely” part). Construed as a purely telic activity, music-making has no value over and above that of having produced a piece of music, and so the more efficiently you can get to that point, the better. AI music generators make perfect sense on this view. The problem, of course, is that the purely telic view of music-making is false. For a musician, the point of making music isn’t merely to end up having made a piece of music: the actual making is part of the poin

That’s not to say that making music is pleasurable all the time. It often involves moments of friction: the high note you’re not hitting, the rhythmic “lock” that just isn’t happening between you and the drummer. But when you push through those moments of friction and get to the other side, what awaits can be some of the most exhilarating, transcendent experiences it’s possible to have as a human being. Those moments more than compensate for the hardship that the friction brings. In fact, they require the friction. Eliminate the friction, and you eliminate the thrill of overcoming it.

651192 511

Stewart Webb @nuclearpidgeon.bsky.social · 16/10/2026

I find that the strengths and weaknesses of LLMs start to make way more sense when you think about them this way

000 0

Stewart Webb @nuclearpidgeon.bsky.social · 16/10/2026

I’ve been saying something much like this for a few years now. “LLMs output a statistically average (plus post-training biasing) completion to their input. Sometimes, the statistical average completion is a useful thing you want. Often it isn’t.”

100 1

Reposted by Stewart Webb

bryan newbold @bnewbold.net · 07/10/2026

ah damn, RIP Margaret Hamilton. she took safety-critical software engineering seriously, and got people to the moon and back. news.mit.edu/2026/margare...

old photos of a woman (Margaret Hamilton) standing next to a tall stack of books (printouts of Apollo space mission software source code)

102144304733 14.4K

Stewart Webb @nuclearpidgeon.bsky.social · 08/10/2026

This thing drives me insane. Such a dark pattern. and especially because it’s on a timer and just locks your whole UI when it comes up!!

010 1

Stewart Webb @nuclearpidgeon.bsky.social · 08/10/2026

Worth watching the Numberphile interview with Tristan Buckmaster who was at the center of that whole mess. He seems pretty convinced they essentially just short circuited most of the way to their result from some kind of lifting from his chat history youtu.be/Mzbtj5nkMXI (and he’s an Aussie!)

youtu.be

The Buckmaster Interview (Navier-Stokes) - Numberphile

YouTube video by Numberphile

111 1

Stewart Webb @nuclearpidgeon.bsky.social · 08/10/2026

I like to imagine that there’s just a giant `if (true)` equivalent hiding in the depths of the Lean proofs or something dumb like that lol

010 1

Stewart Webb @nuclearpidgeon.bsky.social · 06/10/2026

feels like I need a whole browser extension to block his shit out at this point, it infects nearly every programming social media feed I have

000 0

Stewart Webb @nuclearpidgeon.bsky.social · 05/10/2026

That’s cool as. What software is this?

100 1

Stewart Webb @nuclearpidgeon.bsky.social · 04/10/2026

Literally Anthropic in their leaked prospectus: “our models are soooo dangerous they might end humanity. please give us billions of dollars to value our company at $2 trillion”

051 5

Stewart Webb @nuclearpidgeon.bsky.social · 03/10/2026

Only in one particular group effort end result comparative sense here. They are also dumb as rocks in individual scenarios all the time. It’s a broad reductive statement to make without qualifying, which plays into all these stupid god intelligence narratives going around at the moment

000 0

Stewart Webb @nuclearpidgeon.bsky.social · 03/10/2026

Oh for sure. I mean this is where I would somewhat agree and say “the numbers speak for themselves” when it comes to found CVEs etc. But I think it is important to keep the virtual somewhat-limitless scaleup factor in mind as probably the more significant part of the effectiveness here

010 1

Stewart Webb @nuclearpidgeon.bsky.social · 03/10/2026

You can fork 10,000 copies of Claude or whatever with one Kubernetes action on a datacenter for as long as you can power it, but you can’t exactly clone cybersecurity engineers at the drop of a hat and have them all read all of your codebase 24/7 in shifts looking for CVEs

100 1

Stewart Webb @nuclearpidgeon.bsky.social · 03/10/2026

I wouldn’t take this as evidence that it’s “better than humans” per se, it’s just that it can read/analyse so much more code, so many more different ways, 24/7, at such a higher scale than we could have before with however many security-minded humans we have

220 2

Stewart Webb @nuclearpidgeon.bsky.social · 21/09/2026

My current state of mind being in the software industry whilst still doing most of my work myself and using LLMs at most 5-10% of the time:

Simpsons scene of Homer Simpson lying on a couch with a cigar, with the caption "Boy... everyone has psychosis except me."

030 3

Stewart Webb @nuclearpidgeon.bsky.social · 20/09/2026

😂 but poor Larry! How will he cope! (with lawsuits, probably)

110 1

Stewart Webb @nuclearpidgeon.bsky.social · 20/09/2026

Well, still, if anyone had any particular right to re-appropriate the Sun brand… this probably beats Sun merch in an Oracle webshop lol

040 4

Stewart Webb @nuclearpidgeon.bsky.social · 20/09/2026

(My tinfoil hat would go as far to say it’s on purpose because if people realise how stochastic LLM machinery (at least) is, the whole AI illusory facade would collapse… (and yes to be fair Jev sounds like it’s not exactly an LLM))

020 2

Stewart Webb @nuclearpidgeon.bsky.social · 20/09/2026

AI companies produce explainable AI decisionmaking details challenge: impossible

130 3

Stewart Webb @nuclearpidgeon.bsky.social · 20/09/2026

Cosplay? By many accounts Sun was an awesome place to be as an engineer pre-Oracle, and Bryan Cantrill was there on the ground along the way… Oxide has felt to me just like they are trying to keep that good culture going anew

010 1

Stewart Webb @nuclearpidgeon.bsky.social · 18/09/2026

“It used to be, you bought a bunch of computers and shipped em in to a room and turned them on—not anymore! What is going on?”

000 0

Stewart Webb @nuclearpidgeon.bsky.social · 13/09/2026

Sabriel was by an Aussie author?? I had no idea! (Amazing!)

011 1

Reposted by Stewart Webb

Katie Tiedrich @katietiedrich.bsky.social · 09/09/2026

I’m not the first person to make this observation but the prevalence of nauseating ai-generated business logos means stuff like this hits CRAZY for me now

171165813584 16.6K

Stewart Webb @nuclearpidgeon.bsky.social · 08/09/2026

Me trying to avoid using LLMs for programming

040 4

Stewart Webb @nuclearpidgeon.bsky.social · 04/09/2026

It is such a confusing time to be in computer work. People are extolling the virtues of AI everywhere, slop is piling up everywhere, and I can’t decide whether it all leaves my genuine expertise more valuable, or whether businesses are more interested in trying to outsource the need for it entirely

020 2

Stewart Webb @nuclearpidgeon.bsky.social · 02/09/2026

Are there any particular hardware platforms / SoCs this all has best support for? It would be fun to try roll my own seL4/LionsOS system at some point :)

110 1

Stewart Webb @nuclearpidgeon.bsky.social · 01/09/2026

Oh hey, a choir I sing in recently recorded a new composition that set this text. It’s an inspiringly restful set of words for sure youtu.be/Qp00m6lbc4E (Australian composer Joseph Twist)

youtu.be

The Peace of Wild Things

YouTube video by Polyphonic Voices - Topic

010 1

Stewart Webb @nuclearpidgeon.bsky.social · 01/09/2026

Congrats!! I would be super proud to have the actual job title of “lecturer” and you should be proud too 😇

010 1

Stewart Webb @nuclearpidgeon.bsky.social · 31/08/2026

A mistake we learnt a lot from. But can’t leverage the learnings of if we keep Doing More Java

010 1

Stewart Webb @nuclearpidgeon.bsky.social · 21/08/2026

Incredible. I can only imagine how great it would be to see steam tooting out that worried little porg mouth

040 4

Stewart Webb @nuclearpidgeon.bsky.social · 12/08/2026

Are there any services out there as part of Bluesky or AtProto that provide any kind of “snapshots” you can grab to save having to backfill (presumably, every single PDS) the entire way?

120 2

Stewart Webb @nuclearpidgeon.bsky.social · 06/08/2026

Felonies so hot right now in big tech

140 4

Stewart Webb @nuclearpidgeon.bsky.social · 04/08/2026

Oh awesome! I have a bunch of assorted Lilypond projects, I will definitely have to try out sticking them in this kind of web context…

110 1

Reposted by Stewart Webb

Nicky 2 @nickyflowers.com · 04/08/2026

my latest masterpiece, Laptop Smoking a Cigarette (gouache, colored pencil)

a tiny square canvas on a little wooden easel, and the painting? well, it's perfect

236786 367