morphillogical

[morphillogical]

Ponk

Change a single letter and change the word game

I want to play a game with you all.

You have to make a new word by changing only one letter of the last word.

Dirt

[morphillogical]

You can actually only call it hubris if the gods/consequences ever catch you. Otherwise it's just sparkling genius.

[morphillogical]

An autorashoman!

Today in "arguments that people make against a thing that make it sound rad as hell":

I was watching a video on AI art, not really worth my time even at double speed, but they said "can you imagine a movie where you can just position the camera anywhere you want?"

And yeah, I can, sounds fucking awesome.

I mean, the argument was about how a director and editor makes choices, how there's a language to film, the way that a dolly zoom makes us feel and communicates something to us, all that stuff.

But yes, I absolutely would like to watch a movie where I could (virtually) fly around and see things from another point of view. No, it would not have the intentionality, but come on, doesn't this sound like a great idea, even if just as a one-time gimmick? Wouldn't it be great to have people sharing their personal experiences of the same immersive piece of media? How can someone propose something like that while shaking their head sadly about the state of the world?

[morphillogical]

I think part of the dynamic here is that it can be helpful for political coalitions to be ambiguous and less specific. A general idea can attract many different people's interpretations that they can project onto it. A more specific idea is easier to criticize.

So asking for clarification of what, exactly, a slogan or vibes-based rallying cry actually means is partly asking "hey, make this easier to attack while splitting your coalition."

This is one of those "that which can be destroyed by the truth, should be" cases though. It's a very bad sign when a coalition can't stick together when their ideas are made clear.

One thing I find fascinating is ideas where if you ask people what the idea means, they get really frustrated and angry, and say,

"Come on, you can't possibly be this ignorant, it is completely obvious, everybody knows what this means, no thinking person should need an explanation"

And then you say,

"Okay, but bear with me and pretend I'm dumb enough to need an explanation"

At which point you hear multiple radically different definitions, to the point where some are mutually exclusive with each other.

What's fun is that often the people who have radical internal disagreements about some idea are still usually mainly angry at me for pointing it out.

[morphillogical]

This is fairly strongly worded, given that you did only a little better than chance (2/3 correct).

Still thinking about that Astral Codex Ten AI Art Turing test...

I mean... Obviously the one on the right is the human one. Is this some kind of prank? Am I on candid camera?

My suspicion is that what this test demonstrates most conclusively is that we are so thoroughly bombarded with images that we have developed the defensive measure of paying as little attention to them as possible.

We get the gist and then move on as quickly as possible.

Here's someone who did much better than I did on this test explaining their results.

This demonstrates fairly conclusively that nearly all the AI images Alexander chose do in fact, have "tells" which are extremely plain when you attend closely to the details.

In fact, I managed to get 2 out of every 3 correct even with an incredibly lazy and fast-paced assessment carried out on my phone without much recourse to fine detail.

There are two trends I noticed in the comments of the results post.

First, a significant number of ACX posters harbor a suspicion and resentment towards art and good taste, which leads them to suspect that all artistic judgement is essentially arbitrary and based on clout. They don't notice the difference, so there must not be a difference.

Second, a number of people who are clearly AI skeptics gave ground and accepted the idea that the AI images were lacking in "tells" and were especially good, and instead attempted to attack the test on the grounds that this kind of curation was itself unfair.

Both responses indicate, to me, both a fascination with images and a kind of, for lack of a better word, illiteracy about them.

And perhaps most interestingly this illiteracy doesn't seem to obviously vary between pro and anti-AI readers.

To go back to the side by side landscapes up there, the landscape on the left probably has the fewest obvious "tells" of AI art, maybe of all the AI images.

It's also just, you know, a much worse piece of art than the one on the right?

To go back to what I said in an earlier post, the painting on the right draws the eye down the hill. The two figures on the path are expertly set off so that even though they are barely suggested with just a couple of brush strokes, they immediately stand out and draw the eye, causing you to follow the same path they are taking down into the village.

Contrast the image on the left. Which part of the painting is your eye drawn to first? It could really by almost anywhere. No part of the picture is more important than any other, there's very little contrast between, say, the village on the right and the wildflowers on the left. What detail there is is largely because, well, otherwise there wouldn't be a painting.

If you asked 100 art critics which of those paintings was by a renowned master and which one you found hanging in a dentist's office I think all 100 would give you the same answer.

Or take this one:

If you really, really zoom in on the hand on our right, the anatomy is probably wonky, but I didn't notice that, I just thought,

"Okay, but, like, what is this angel, like... Doing?"

This figure, painted in this style, is rife with symbolism. Most likely an angel, or at the very least Icarus, it ought to be extremely clear what sort of emotional/cultural/allegorical/etc. meaning is being communicated, but it is just sort of... looking off yearningly towards nothing.

Culturally, it's just not something that a human would paint as a finished piece.

Actually in general AI seems to tend to either not have a clear focal point, or to have one extremely obvious subject placed right smack dab in the center of the frame.

One of the subtle visual gags in Monty Python and The Holy Grail is that the peasants are often doing things that look, on very cursory examination, as though they are some kind of agricultural activity, but actually they are just hitting random patches of ground with a stick or sitting on the ground and moving mud into a big pile.

And same with this Angel; it looks, at casual glance, to be doing "Angel type stuff" and if you just keep moving you leave with the impression that everything was fine.

But if you stop yourself, go back, and ask, "Wait, specifically what is it doing?" you really can't come up with anything more specific than, "Angel type stuff".

This sort of vagueness is also a tell of AI art.

If what I'm saying sounds a bit frustrated or mean-spirited I think it's because looking at this test has solidified something that I haven't really been able to articulate before, which sort of sums up to the vast majority of talk about AI, regardless of what the conclusion is, evidences a strong emotional investment in images paradoxically combined with a sort of estrangement from them and often even a strong resentment towards them.

Both pro and anti-AI imagery camps contain a tremendous number of people who feel imagery as a kind of imposition, with AI as either an emancipatory force aimed at a tyrannical art world bent on crushing us with arbitrary, incomprehensible images or, on the other hand, as a tyrannical force set to flood us helplessly with a set of incomprehensible images almost entirely against our will.

[morphillogical]

this is just a Quentin Tarantino character who didn't realize he was supposed to be fictional

I still can't believe that Tommy "Karate" Pitera, the mafia boss who was also a karate champion and who killed people by doing karate on them, was a real person. That's not a combination of traits a real guy has. That is a Metal Gear character. That is Chuck Norris playing the heel in a Bruce Lee movie. But no, there was a time not that long ago when people on the mean streets of Brooklyn might whisper "Hey, watch out for Tommy Karate, he'll kill you with his bare hands" with the gravest seriousness.

[morphillogical]

WAIT I FORGOT ABOUT YUGIOH UWU DAY SHIT

[morphillogical]

gal who's about to invent Las Vegas

it would be easier to be a tourist if they didn't put the tourist traps so far apart. If they just put them in one place, you wouldn't need to do some much travelling to get to them.

[morphillogical]

I'll be honest, I needed to be told this

Don’t give CPR to wild animals: a thing you wouldn’t think you’d have to tell people but apparently you do. 💀

[morphillogical]

I've been saying "Gadzooks" once I learned the incredibly hardcore thing that it's short for

guy who says "FUCK!" to every minor inconvenience x guy who says "oopsie daisies" to earth shattering catastrophes

[morphillogical]

But None isn't a language primitive it's part of an ordinary type in the standard library. That's unusual!

[morphillogical]

runs in browser

link friend to app, they click, any device, no install, no gatekeeper, just works.

I put up with a lot for this. I'd put up with a lot more

[morphillogical]

Shapeshifters stay winning.

All the shrimp colors, none of the hell dimensions, plus you get to turn into a little mouse sometimes, as a treat.

Thank you @nitewrighter for your post lmao

[morphillogical]

you can shake him. he enjoys it

[morphillogical]

Hm, this is cool, but unless it was followed shortly by all the zombies becoming skeletons, I think it would make it much harder for me to sustain my disbelief.

Because those huge clouds of flies are eating the zombies right? Or their maggots are? And when something is eaten, it's not part of the original thing anymore. That's one of the big things about eating.

It'd work in a medium that's very zoomed out, and/or cartoony. Like how some settings have zombies that tear off parts of their body to throw as a projectile. But a setting that's more realistic, like a 28 Days Later or a Walking Dead it'd be harder to ignore.

Funny how we draw the lines though. I mean, we've all collectively agreed not to think about where zombies are getting their energy and nutrients from. Oh, or maybe the thinking is that they're like, cold blooded and dry like a lizard? That would alleviate the problem a bit, letting them go fairly long periods of time with little food.

But maggots eat flesh, and if you've got huge clouds of flies for miles around, well, there ain't gonna be much flesh in them there corpses for long.

Would be cool to see a zombie setting switch to a skeletons setting. Maybe the skeletons could still spread the zombie plague? That'd be cool.

If zombies were real, you wouldn't first be warned by the approaching horde by their smell, by their groans, not even a cloud of smoke of the dust they raise coming closer from the horizon. It would be the flies. Hordes and hordes of insects, corpse-flies laying eggs on the carcasses of people who still walk, eating the eyeballs from their sockets, climbing across their unfeeling leathery skin. And the buzzing. The inescapable, deafening buzzing. Everywhere. Like you did not just kick a hornet's nest, but the very ground you walk on was a hornet nest, and each step caused another explosion of insects.

Insects, corpse flies, the buzzing. Their swarms blacken the skies, more horrifying than their migrating meals. The deafening cacophony of constant buzzing, the horrid noise of the living who feast on the dead who feast on the living. The buzzing.

The endless, inescapable buzzing.

[morphillogical]

Ah! Yes, this I endorse.

Many people building things that don't work today. Generously, they're skating to where the puck is going, which is hard to do when it's moving in such erratic bursts. In that case, they're going to have to be uncharacteristically patient to not release something that won't work for a generation or two of models.

Less generously, a lot of people are dreaming more than looking, and wishing more than thinking. That group working on health products is s pretty terrifying thought lol

I feel like the AI risk rats got it almost exactly backwards actually.

[morphillogical]

Disagree. GPT-N was trained on next token prediction not to make chatbots, but because it's a large corpus that lives in the shadow cast by human thought. OpenAI didn't even deploy a chatbot interface for it until the end of 2022, >5 years from the transformers paper and >4 years from the first GPT paper.

They're very explicitly trying to make AGI with it, and they have made significant progress towards that goal. Cutting edge models have a degree of general intelligence that would be astonishing to any AI researcher before ~2020.

I feel like the AI risk rats got it almost exactly backwards actually.