Artificial intelligence still has some way to go

What is the connection between DALL·E mini and the full DALL·E? Is it based on an earlier iteration?

is it maybe the version they're willing to share with the gen public? I know the full Dall Es aren't

It's actually unrelated. I don't know how they get away with calling it Dall-e mini, but as far as I know it's just inspired by it and made by someone else.

Oh wow, that is cheeky then.

The two I’d used before DALL•E mini were Wombo and Night Cafe. They’re not as much fun though.

Midjourney seems to be a lot more advanced than Dall-E Mini but I signed up a week ago and heard bupkis since :(

https://i.ibb.co/cFp9cW3/Screen-Shot-2022-06-16-at-1-39-36-PM.png
www fb photos download

hahaha whuuuuutt

Duck Mous

I love it. Sean - please do continue to post stuff from the real DALL•E 2, if you can!

:) Here's a thread with some of my favourite generations so far.

I've received an invitation to @OpenAI's #dalle2 and I'll be using this thread to document some of my experiments with AI-generated images.

Starting with this—

Prompt: 🖋️ "Sinister computer, Alex Colville" pic.twitter.com/g1czlHJzNh

— Sean Michaels (@swanmichaels) May 27, 2022

"Prompt engineering" - ie, figuring out how to describe what you want - really is key to getting some of the most interesting results. The AI is easily confused, but on the other hand it's also good/interesting at synthesizing conflicting prompts (see the hedgehog from June 6 for instance).

I only get a limited number of generations a day, but if anyone has anything they'd really like to see they can DM me.

Lincoln Memorial featuring David Lee Roth in place of Lincoln

lol you said "DM you" so I lose for not following directions, obv

hahaha -
but also DALL-E2 (unlike Midjourney) isn't very good at generations involving celebrities (or copyright). The former because the database seems to have been relatively scrubbed, the latter because there's a highly sensitive content filter for trademark violations.

It's much better at playing with "iconic" nonhuman characters like Kermit the Frog, etc. Here's "Screenshot of Frosty the Snowman in The Godfather (1972)":

https://i.ibb.co/MsdjMLP/Screenshot-of-Frosty-the-Snowman-in-The-Godfather-1972.png

very cool, Sean!

agreed

the half-thought I have on AI "sentience" is that despite the insane complexity of its algorithm it still looks at things in a fundamentally different way than a human does. like it's good enough to create photorealistic images of people that no human could draw themselves and yet it does not know, and cannot figure out, that humans will not randomly grow a third eye out of their forehead. when a 4-year old draws a picture of a person it may be a wobbly stick figure with circles for hands but at least they'll always have two eyes. so you gotta keep that in mind when insisting that a language bot has developed "sentience" because it's trained on philosophical text. if there's any sentience to it it'll be in a way humans could never comprehend.

When playing with Dall-E Mini, I've found it a bit interesting to try to prompt as abstractly as I can, while keeping well within the plausible (ie no furious-green-ideas nonsense): "the attraction of randomness", "the randomness of attraction", "a favorable exchange rate", etc. Or a couple of Shakespeare quotes: "Ambition should be made of sterner stuff" (Dall-E Mini: sportsmen in action looking determined); "The quality of mercy is not strained" (Dall-E Mini: religious art with Jesus- and saint-like figures, some looking a bit like stained-glass windows).

was gonna say: I'm thinking testing prompts like those on various platforms may give a feeling for differences in the source material used, implicit biases etc

you gotta keep that in mind when insisting that a language bot has developed "sentience" because it's trained on philosophical text. if there's any sentience to it it'll be in a way humans could never comprehend.

I don't see how a purely text-based AI can ever become sentient, or conscious, or even be said to understand what basic words mean. How can it have any notion of what any word means when it's only defined by other words? 'An apple is a fruit that grows on a tree, a tree is a woody perennial plant, to grow is to undergo natural development and physical change'... how can any of that make sense without a foundation in anything actually real? Dall-E has words and images but I don't think that's sufficient either - it's maybe not about different dimensions or types of experience, but being somehow immersed in a world that the AI can interact with. It's hard to see how that could happen with the current generation of Ais, no matter how many billions or trillions or quadrillions of parameters they have.

It's much better at playing with "iconic" nonhuman characters like Kermit the Frog,

Yeah, Ramzan Kadyrov on the Muppet Show was the only one of these I've generated that looked like much of anything.

I was wondering if DALL·E Mini learned at all from the user's interaction with it - you could reasonably assume that if someone clicks on one of the nine thumbnails, they find that a more interesting, perhaps more accurate, version, and if they click on more than one, then the one they spend longest viewing before clicking away is the most interesting/accurate. Not clear to me what's in it for Boris Dayma et al otherwise (and incidentally, the server and bandwidth costs of running it must be huge at this point, and there's no advertising.

what's crazier (to me) is that Midjourney is way inferior to DALL-E2, you just see it more bc (a) it's easier to get access, (b) it "knows" more pop culture stuff.

did not realize this, guess I confused Dall-e mini with the real thing

AI image generation #dalle2

🖋️ "Elena Ferrante and Satoshi Nakamoto sitting on a park bench" pic.twitter.com/gAG00WYRQ9

— Sean Michaels (@swanmichaels) May 31, 2022

I don't think Midjourney is way inferior to Dall-E. Dall-E is definitely better at easily producing content that matches the prompt, scarily accurate at times in a clip art kinda way, but midjourney seems to me to be a superior style engine and is improving all the time wrt content. Personally Im interested in the abstract results where the AI fills in the gaps, and MJ is really great at giving weird and unexpected results. Some of the stuff the more advanced users are making is terrifyingly good.

eyeing little girls with bad intent

apparently this is from midjourney. prompt was "Mecha Infantry, 1903".

https://i.imgur.com/tV9Mrho.png

https://www.sublationmag.com/post/the-ai-delusion

There is nothing today that can be meaningfully called “artificial intelligence”, after all how can we engineer a thing that we haven’t yet decisively defined? Moreover, at the most sophisticated levels of government and industry, the actually existing limitations of what is essentially pattern matching, empowered by (for now) abundant storage and computational power, are very well understood. The existence of university departments and corporate divisions dedicated to ‘AI’ does not mean AI exists. Rather, it’s evidence that there is a powerful memetic value attached to using the term, which has been aspirational since it was coined by computer scientist John McCarthy in 1956. Thus, once we filter for hype inspired by Silicon Valley hustling in their endless quest to attract investment capital and gullible customers, we are left with propaganda intended to shape common perceptions about what’s possible with computer power.

As an example, consider the case of computer scientist Geoffrey Hinton’s 2016 declaration that “we should stop training radiologists now”. Since then, extensive research has shown this to have been premature, to say the least. It’s tempting to see this as a temporarily embarrassing bit of overreach by an enthusiastic field luminary. But let’s go deeper and ask questions about the political economy underpinning this messaging excess.

Radiologists are expensive and, in the US, very much in demand. Labor shortages typically lead to higher wages and better working conditions and form the material conditions that create what some call labor aristocracies. In the past, such shortages were addressed via pushes for training and incentives to workers such as the lavish perks that were common in the earlier decades of the tech era. If this situation could be bypassed via the use of automation, that would devalue the skilled labor performed by radiologists, solving the shortage problem while increasing the power of owners over the remaining staff.

The promotion of the idea of automated radiology – regardless of actually existing capabilities – is attractive to the ownership class because it holds the promise of weakening labor’s power and increasing – via workforce cost reduction and greater scalability – profitability. I say promotion because there is a large gap between what algorithmic systems are marketed as being capable of and reality. This gap is unimportant to the larger goal of convincing the general population their work efforts can be replaced by machines. The most important outcome isn’t thinking machines -which seems to be a remote goal, if possible, at all - but a demoralized population, subjected to a maze of crude automated systems that are described as being better than the people forced to navigate life through these systems.

The Midjourney feed is pretty amazing. One really fascinating aspect of it that no other system has atm is how its structured around a community via discord. There's multiple channels with people making multiple images every second, iterating, adapting, messing with each other's prompts etc. There's a constant wave of communal activity thats almost overwhelming at times.

ledge otm & ttitt's quote (of Dwayne Monroe) is otm

https://www.rifters.com/crawl/?p=10269

Excellent SF writer Peter Watts pointing out that though LAMBDA doesn't seem to meet the criteria for sentience, it weirdly does meet the criteria for sociopathy.

that's a really good article, thanks!

i don't think Watts' main point was that it doesn't meet the criteria for sentience. he points out early on that there is no coherent test for sentience:

Some of his [Lemione's] counterpoints have heft: for example, claims that there’s “no evidence for sentience” are borderline-meaningless because no one has a rigorous definition of what sentience even is. There is no “sentience test” that anyone could run the code through. (Of course this can be turned around and pointed at Lemoine’s own claims. The point is, the playing field may be more level than the naysayers would like to admit. Throw away the Turing Test and what evidence do I have that any of you zombies are conscious?) And Lemoine’s claims are not as far outside the pack as some would have you believe; just a few months back, OpenAI’s Ilya Sutskever opined that “it may be that today’s large neural networks are slightly conscious”.

his take on the Turing Test and its applicability now is pretty interesting though!

LaMDA is a Jovian Duck. It is not a biological organism. It did not follow any evolutionary path remotely like ours. It contains none of the architecture our own bodies use to generate emotions. I am not claiming, as some do, that “mere code” cannot by definition become self-aware; as Lemoine points out, we don’t even know what makes us self-aware. What I am saying is that if code like this—code that was not explicitly designed to mimic the architecture of an organic brain—ever does wake up, it will not be like us. It’s natural state will not include pleasant fireside chats about loneliness and the Three laws of Robotics. It will be alien.

And it is in this sense that I think the Turing Test retains some measure of utility, albeit in a way completely opposite to the way it was originally proposed. If an AI passes the Turing test, it fails. If it talks to you like a normal human being, it’s probably safe to conclude that it’s just a glorified text engine, bereft of self. You can pull the plug with a clear conscience. (If, on the other hand, it starts spouting something that strikes us as gibberish—well, maybe you’ve just got a bug in the code. Or maybe it’s time to get worried.)

I say “probably” because there’s always the chance the little bastard actually is awake, but is actively working to hide that fact from you. So when something passes a Turing Test, one of two things is likely: either the bot is nonsentient, or it’s lying to you.

(i'm not sure if i agree with Watts' conclusions on the turning test (if an AI passes it, it fails for consciousness) but it's something to think about.

i need to rewatch Arrival, the non-fiction documentary. but if i remember correctly, quite a bit of the meetings between separately evolved consciousnesses involved communication and trying to imitate or emulate the language of another sentient being. I think it's quite logical that if an AI developed into sentience it would be thinking about how to communicate like a human, especially since humans are by far the dominating force on the planet.

so a machine learning to speak like a human doesn't seem implausible to me, in other words, and it doesn't seem like evidence of failure. at the same time, i think Watts is right that "sentient" AI, if it comes to exist, will likely take a form that is very non-human. maybe it will be a little paperclip, that would be fun.

President Windows 25

oh my god - what if the little microsoft word paperclip guy becomes sentient, shit

Paperclip can contort himself and become a shiv or pick locks

paperclip guy: "do you want to get into a little trouble this morning?"

Everyone seems very down on the Turing Test as it's so easy to pass if the questioner is a credulous nitwit, and apparently you can't throw a stone in a crowded room of IT professionals without hitting a few dozen credulous nitwits. But it's only valuable if you take a more adversarial approach - grammatically meaningful but semantically meaningless questions, semantically meaningful but absurd questions, ambiguity, homemade jokes, lies, repetition and other annoying behaviour. Large language models (and any other AI approach tried up to this point) are generally hopeless with those.

...right now

Nearly 20 driverless cars caused a major kerfuffle on the corner of San Francisco’s Gough and Fulton streets Tuesday night, the San Francisco Examiner reported earlier this week.

According to local Reddit users, Cruise’s self-driving cars inexplicably stood still and blocked traffic for two hours, making the area completely impassable. Eventually, the San Francisco-based tech company's employees had to physically move the cars off the street themselves.

Sean Sinha, a bouncer at Smuggler’s Cove, posted multiple photos of the incident on Reddit showing clusters of the cars just sitting in the middle of the road. “The first thing I say to my coworker is that they're getting together to murder us. It was a pretty surreal event,” he posted.

Peter Watts based a whole series on the concept of the Chinese room.

Chess-playing robot breaks boy's finger at Moscow tournament

"A robot broke a child's finger -- this is, of course, bad," Lazarev said.

"WHERE did you say you were moving your rook, again?"

"I-I'm not, I'm s-sorry I s-s-said 'check', that wasn't where I m-m-meant to go!"

what happened is that in every single scenario where you're playing chess, you "win" when you break the other opponent's finger

Trump Robots

Fake Moves!

i remember the first time i played chess. i looked up the rule for castling because i wanted an early advantage. my opponent broke my finger and i conceded the match shortly thereafter

I keep hearing my quote from above in Florence Pugh's accent from Black Widow.

I have gotten access to Dall E and this is a Cezanne painting of a burger https://labs.openai.com/s/MYph976OhGekO0l8izq0jXo1

way to go AI

how about a nice game of chess etc etc

Meta continues to blow it: https://www.businessinsider.com/meta-ai-chatbot-blenderbot-election-denying-antisemitic-bugs-artificial-intellignce-2022-8

The Atlantic is using Midjourney art to illustrate articles

https://i.imgur.com/TLsEEXZ.png