Random Thread

[malloc]

Just to clarify some things, I am not making absolute deterministic predictions about the future, only suggesting probable forecasts and appropriate preparations. If you see storm clouds gathering on the horizon, it makes sense to get your umbrella or even watch out for lightning and hail. Perhaps the storm will dissipate or move the other way, rendering your preparations pointless, but if not, you will have saved yourself considerable trouble. Pay no attention to the fool waving "An Inquiry Concerning Human Understanding" at you while standing in an open field under a looming thunderhead.

[Travis B.]

malloc, what you are doing is akin to saying "the Nazis are unstoppable, look at how they rolled over Poland, France, the Low Countries, Denmark, and Norway" in 1940. The Nazis weren't beaten by calling them unstoppable. They were beaten by refusing to give in to them even at the darkest hour, when Britain effectively stood alone against Nazi Germany and the Soviet Union (which at that time was on Nazi Germany's side), while America largely stood on the sidelines.

[malloc]

Quite the contrary, I would consider myself rather optimistic for believing that we can change the course of technological development and even uninvent harmful technologies. Nonetheless the first step to that is recognizing the power and danger of those technologies and it feels like many people resist that point. Claiming that AI will stop at coding and mathematics is like someone in 1940 claiming the Nazis will stop at Poland and France and leave everyone else alone.

[Travis B.]

Thing is, what we can do is refuse to give into the AI bros' hype. The AI bros want people to think that AI is all-powerful, that we are almost to the point of AGI, when in fact we are not. That is why they stage things like the 'accidental' hacking of Hugging Face -- even AI disasters only help to hype AI. The more we think AI is all-powerful the more money the AI bros will make when Anthropic and OpenAI go public. Insisting that one is against AI while also saying that AI is all-powerful does not help; it still adds to the AI hype and thus the AI bros' profit when they go public.

[bradrn]

That is why they stage things like the 'accidental' hacking of Hugging Face -- even AI disasters only help to hype AI.

FWIW I see no reason to believe it was staged. (In general, my default position is to reject conspiracy theories.)

[Travis B.]

That is why they stage things like the 'accidental' hacking of Hugging Face -- even AI disasters only help to hype AI.

FWIW I see no reason to believe it was staged. (In general, my default position is to reject conspiracy theories.)

They have openly admitted that they deliberately let it happen by removing the 'don't hack other people' guardrails for the purpose of a test.

[Travis B.]

The key thing is that they very openly admitted that their AI did it, whereas most corporate disasters are carefully hidden whenever the least possible -- and this is because AI disasters like this one help the AI bros by portraying the image that their AI is so powerful.

[bradrn]

That is why they stage things like the 'accidental' hacking of Hugging Face -- even AI disasters only help to hype AI.

FWIW I see no reason to believe it was staged. (In general, my default position is to reject conspiracy theories.)

They have openly admitted that they deliberately let it happen by removing the 'don't hack other people' guardrails for the purpose of a test.

That’s like saying that, because I kept some raisins in a non-airtight box, I wanted my room to have a fruit fly infestation. (A real incident that happened the other day. I learnt my lesson and am looking for a better jar now.)

More generally, creating the preconditions for a certain outcome is not the same thing as wanting or expecting that particular outcome. This is the whole reason why ‘unintended consequences’ are such a common thing. In this case, by removing the guardrails, the testers expected the LLM to try hacking software within its carefully prepared sandbox — not to try hack through the sandbox itself.

The key thing is that they very openly admitted that their AI did it, whereas most corporate disasters are carefully hidden whenever the least possible -- and this is because AI disasters like this one help the AI bros by portraying the image that their AI is so powerful.

This logic suggests that it’s in their interests to openly admit it no matter whether it was staged or not.

[Travis B.]

Apparently, though, according to this OpenAI's AI mostly used techniques that human crackers would use, and was very noisy in what it did because it had no impetus to try to be stealthy at all, whereas most human crackers would have been much quieter in what they would do. It could have been stopped by Hugging Face had Hugging Face had basic security measures like defense-in-depth in place, so it is not like the AI got in through some kind of all-powerful cracking ability. The main thing notable about the AI, if anything, was how aggressive it was without caring about whether it was detected or not.

[alice]

How long will it be before AIs start hacking into each other and the whole thing comes crashing down?

[zompist]

Zompist effectively does disagree, though, given that he doubts the current trajectory will hold much longer and considers that reason enough not to worry about AI getting too powerful.
[...]
He has consistently dismissed my concerns about AI continuing to advance as the fallacy of induction. He once compared them to expecting modes of transportation to continually improve until everyone has hypersonic vehicles. While he does not necessarily rule out the possibility that AI will continue to improve, he evidently considers it unlikely enough that it doesn't make sense to develop active policies against it.

Based on his past comments, it sounds like he has lived through multiple waves of AI enthusiasm that all ended in failure, leaving him with inherent skepticism toward the field. Fair enough, but that doesn't change the fact that current AIs are clearly working in ways that previous ones could not.

More complete lies. I'm tired of it. I'm sorry that you feel the need to create a shadow-zompist to get mad about, but you don't get to do it here. For now you are banned for a week, next time will be permanent.

[Man in Space]

This may be an unusual inquiry.

I have, for some time, liked to create album art for nonexistent albums. This allowed me to achieve multiple creative aims at once: I've always loved fonts and typefaces but had nowhere to use them; certain "artist" names, "album" titles, and pairs thereof simply sound too good not to exist; it allows me to make music-related art in other ways; I get to practice my editing and layout skills. Our board's own cartilaginous comrade enjoys making banknotes and in-universe documents; however, as Dr. Shark's banknotes are consistent with/part of his greater authorial universe whereas my album covers are just one-off ideas with a common given form, I think his images have more gravitas in them than mine would. If I would be permitted to share some of these on the board, what would be the proper venue for it? I don't know if Conlangery is appropriate since these are just purely to allow me the joy of creative service and aren't part of a coherent, overall whole or project, and I don't know if this warrants a whole separate thread--especially given that, in practice, a scattershot schedule of actual posts is anticipated (who knows when an idea will strike?).

[lëtzeshark]

This may be an unusual inquiry.

I have, for some time, liked to create album art for nonexistent albums. This allowed me to achieve multiple creative aims at once: I've always loved fonts and typefaces but had nowhere to use them; certain "artist" names, "album" titles, and pairs thereof simply sound too good not to exist; it allows me to make music-related art in other ways; I get to practice my editing and layout skills. Our board's own cartilaginous comrade enjoys making banknotes and in-universe documents; however, as Dr. Shark's banknotes are consistent with/part of his greater authorial universe whereas my album covers are just one-off ideas with a common given form, I think his images have more gravitas in them than mine would. If I would be permitted to share some of these on the board, what would be the proper venue for it? I don't know if Conlangery is appropriate since these are just purely to allow me the joy of creative service and aren't part of a coherent, overall whole or project, and I don't know if this warrants a whole separate thread--especially given that, in practice, a scattershot schedule of actual posts is anticipated (who knows when an idea will strike?).

Go for it! There is, I think, a thread called the "Daily Creativity Thread", but they could also easily fall into conworlding if you believe enough in yourself. And thanks for the flattery. :P

More: show

Incidentally and on a semi-related note, I also did a joke album cover once for my brother. (Long story. Yes, that's my actual handwriting.)

GGHIV-jpg.jpg (185.85 KiB) Viewed 4170 times

[bradrn]

I like your handwriting! Though I’m a little confused as to why that squiggle is read as ⟨G⟩…

[lëtzeshark]

I like your handwriting! Though I’m a little confused as to why that squiggle is read as ⟨G⟩…

Thanks! And here's how that became my ⟨G⟩ (though I use more version 2.0 when I'm writing out Christmas cards):

IMG_9168 Small.jpeg (28.26 KiB) Viewed 4164 times

[bradrn]

I like your handwriting! Though I’m a little confused as to why that squiggle is read as ⟨G⟩…

Thanks! And here's how that became my ⟨G⟩ (though I use more version 2.0 when I'm writing out Christmas cards):

Ah, I see. My ⟨G⟩ looks like the one in the middle. The US shape is unfamiliar to me.

[lëtzeshark]

I like your handwriting! Though I’m a little confused as to why that squiggle is read as ⟨G⟩…

Thanks! And here's how that became my ⟨G⟩ (though I use more version 2.0 when I'm writing out Christmas cards):

Ah, I see. My ⟨G⟩ looks like the one in the middle. The US shape is unfamiliar to me.

The one on the left is what they at least normally taught in elementary school back in my day: the US traditionally used versions of the Palmer Method, such as D'Nealian (what I was taught in second grade) or Zaner-Bloser (what they switched to when I was in third grade for some reason). That said, because I grew up in Germany (kinda), I was also exposed a bit to Ausgangsschrift (which I found cooler in some respects), so some of my letters became a hodgepodge of the styles. (And things have continued to change over the years as well!)

[Raphael]

Going back to "AI" for a moment, I found this brief thread about the possible future of "AI" generated "art" on Bluesky:

https://bsky.app/profile/opinionhaver.b ... ykb5nxd22g

I’d bet:for a while the cordon sanitare holds & reputational penalty for AI use means the the only people who ‘get away’ with using it will be skilled in it’s use (either hiding it or the end product is good enough no one cares), this causes floodgates to open eventually, then the slopocalypse

There will be bad acts that get panned and don’t find an audience, then a couple of breakthroughs doing geniunely creative and compelling things with it that will find an audience regardless of critics, then you’ll have a generation of audiences and creatives that are fans of them, then all is lost

Basically the strident critics of AI in art will be at first seemingly vindicated, then discredits, then disregarded, and then ultimately totally vindicated about 15-25 years after they thought they would be when the actual really bad slop arrives

[zompist]

Going back to "AI" for a moment, I found this brief thread about the possible future of "AI" generated "art" on Bluesky:

Seems incoherent. Does this person think AI art will be "really bad", or so good that it's "geniunely creative and compelling"?

All these outcomes seem possible to me (not necessarily all at once):

* AI needs another big breakthrough to get past its current problems. That is, "add more inputs" and "add more weights" don't work forever.
* AI pollutes the web so much that it's increasingly difficult to get good training sets.
* AI turns out to be too expensive to (say) make movies, once people have to pay its real price.
* AI crashes so hard that big business moves on to some new fad.
* AI is quietly used much like CGI: for upscaling, backgrounds, concept art, special effects.
* Some crappy segments of the market can be done by AI. There's already a market for things like clip art and elevator music.
* Some people will make a weird melange of human art with heavy AI help. Like, a comic strip with photorealistic art because the creator can't draw.
* The uncanny valley doesn't end. People learn to see the tells. Heck, maybe AI evolves enough that it can reliably detect AI.

I don't think AI will ever take over art and literature in a mallocian sense, even if it gets harder to detect. That's because I, and millions of other people, want to see what humans think and come up with, want to see musicians playing musical instruments, want things that are new instead of pastiche, want precisely the hesitancies and imperfections that come with human art. Plus, humans will continue to want to create art.

Maybe this is a minority taste, but I wouldn't be so sure. There were predictions that e-books would destroy print books, and they haven't. And this isn't some sort of elitism. E.g. in the fanfic world, there are a lot of people who strongly disapprove of AI fanfic.

[alice]

I think a pertinent comment comes from, of all people, Frodo Baggins: "The Shadow can only mock, it cannot make".