Notices by James Widman (jameswidman@mastodon.social), page 2

unhelpful/misleading term: "train" as applied to LLMs

better terms: "compile", "build"

there's still people who cite the founders of american government as if they're role models.

here's a thing about argument from authority: it can be reasonable when the "authority" has some special expertise or experience. but men like jefferson, washington, etc were _explicitly_ white supremacists. so they never developed any experience in designing a government that attempts to treat everyone fairly.

so if you're going to argue from authority, then you'll need to find better authorities.

@GeoffWozniak @thomasfuchs that and a nice walk or a shower.

also, sleep. the power of sleep is unmatched.

and it's also nice if there's a person who's willing to be your rubber duck.

@futurebird @SnoopJ ok, so in the chatbot source code, there must be some point, after it appended the user's latest chunk of text to the end of the document, and _before_ it calls some "inference" function, where it appends a string to the end of the document like:

"</forum post><forum post> ChatbotUser:"

this makes me think that, for a large number of users, the spell might break if we could show how the chatbot behaves when we comment out that little string appendage.

@futurebird @SnoopJ
(and it also might help if we could have a full debug view of the ongoing document, which might contain lots of other annotations like this that help explain how a next-token-predictor could produce something that fools us into thinking that it could only have been written by a feeling & sentient being.)

RE: https://hachyderm.io/@dalias/116579073318810628

perfectly good tram lines were torn out of u.s. cities in order to induce demand for cars and oil.

the auto & oil industries as we know them are mostly make-work. it's unfathomable waste, fraud, and abuse.

there ought to be* some kind of study on the amount & severity of harm done as a consequence of men making decisions on the basis of whether their egos were inflated or challenged.

[*] links/references welcome

if the past is a different country, then so is the future.

a thing about "AI is here to stay" arguments:

@RogerBW @futurebird related: people cheering on the iranian military after they hit AWS data centers:
https://bsky.app/profile/richard.wickedproblems.earth/post/3mgf7hc3vcs2s/quotes

like, they're not only noticing that it is valid & legal to target enemy data centers during wartime; there's also tons of people writing words to the effect of "dear iran, please hit more of our data centers."

@RogerBW @futurebird i used to think i was anti-war, but really, i was always just anti-aggressor.

When you live in the aggressor country, being anti-aggressor usually translates to political demands to stop fighting. This was true for the u.s.-initiated war in Vietnam; the u.s.-initiated war in Iraq; the 2nd u.s.-initiated war in Iraq; etc.

But this time, i don't just want the fighting to stop; i also want the u.s. military to lose, because a "victory" would mean that evil has won.

@nick @mcc thankfully the swift project's fork of LLVM is clean for now:

(or at least, it hasn't been touched by claude yet)
https://github.com/swiftlang/llvm-project

@inthehands in the near-to-medium term, i'm on board with this.

for the longer term, it seems like understanding that reluctance (and knowing how to deal with it) could be very important in preventing the rise of other fascist regimes. So at *some* point, there ought to be some kind of postmortem analysis.

@eniko @cwebber yeah; the addiction part is where i start to feel some sympathy.

Like, probably most people can play a few rounds of a tower-defense-style game, put it down, and say, "well, that's enough for me!", whereas i will say, "i'll put it down in a few minutes; i just want to do one more round."

Then like 500 rounds pass, during which time i barely eat or sleep. "Just one more round, bro."

And that's why i haven't touched a tower-defense game in years.

@whitequark I expect everyone hitting the "favorite" button is feeling stunned by how awful this situation is...

There's way too many people who need to learn that, sometimes, friendship is an exclusive-or type thing; and that if we fail to acknowledge this, we can enable much, much more harm than we expect. And this applies much more often than we want to admit.

I hope more people in your life will show that they understand this.

And I'm glad you're still here.

@whitequark @stella @ariadne
but "because it feels off-mission", in contrast, sounds like something much more fluid. It's not coming from a place of principle, but from a sense of what *feels* expedient to him, **for now**, for a set of circumstances that could potentially change.

@whitequark @stella @ariadne yeah; like, even if we take the charitable read where the reporter *asked* about it first (instead of Enzor-DeMeo just saying it unprompted), he could have responded:

"Out of the question. It would make millions of users significantly more vulnerable to malware. Why would you even ask me that?"

or even just:

"we would lose a lot of users if we did that."

...

@maco @dalias The innovative part that they take 30% of every sale—y'know, like a credit card company, but more so. The innovation is in the "more so".

@whitequark @hailey but i want to live

@whitequark you mean explanations for questions like, "why is that presenter standing within a few feet of a pair of animals in a state of blood lust when he found out that each of them weighs as much as a ford mustang?"