Artificial intelligence still has some way to go

(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

:/

That's actually quite interesting, because it's a passable enough illustration. I can imagine content generators using Craiyon not as a novelty, but as a legitimate way to illustrate articles with an endless supply of royalty-free images. It's almost tailor-made for something like The New Yorker, where there are little cartoons all over the place. If I was a professional cartoonist I would be nervous, because this kind of thing can't be un-learned, and it's only going to get better.

If the caption hadn't credited Midjourney the picture would look like a clever modern cultural reference to Midjourney, and then over time it would just be an illustration like any other. It raises the question of whether illustrators are already passing off its work as their own. Or whether the CIA used an early version of the same technology to make Ren and Stimpy, and John Kricfalusi isn't actually real. Or he's a normal man who has been drugged and brainwashed into acting like a nutcase. By the CIA.

For the record, if I feed Craiyon with "Alex Jones inside an American Office under fluorescent lights" I get a little pop-up message that says "So you're saying that South America doesn't exist, is that right? What about Venezuela, is that not America? There's more to America than hot dogs and Donald Trump. Educate yourself! A luta continua" and then it refuses to go any further. Because I wrote "America" instead of "the United States".

It angers me that I have to write "people from the US" and "the US audience" and "internet users in the US" instead of "Americans", "Americans", and "Americans", whereas Charlie Warzel - whose name sounds like a child gargling - can write "American" and no-one cares. He gets a free pass because he's one of us. It angers me.

Our AI technology can now clone anyone into a 'Digital Human' avatar.

And... we can now clone voices too. Even difficult ones, like @StephenFry.

Give it a go, and talk to him here: https://t.co/5Jf64mCrE3#digitalhuman #virtualbeings #interactiveavatar pic.twitter.com/fZ0aqvTAY1

— trulience 🤖 (@trulience) August 23, 2022

that kind of sucks? he sounds very computery and looks like he's in a PS4 game

Yeah, was bringing the thread back on topic ha ha

mission accomplished then.

https://arstechnica.com/information-technology/2022/08/ai-wins-state-fair-art-contest-annoys-humans/

A synthetic media artist named Jason Allen entered AI-generated artwork into the Colorado State Fair fine arts competition and announced last week that he won first place in the Digital Arts/Digitally Manipulated Photography category, Vice reported Wednesday based on a viral tweet.

Allen used Midjourney—a commercial image synthesis model available through a Discord server—to create a series of three images. He then upscaled them, printed them on canvas, and submitted them to the competition in early August. To his delight, one of the images (titled Théåtre D'opéra Spatial) captured the top prize, and he posted about his victory on the Midjourney Discord server on Friday.

There are several factors at work there to make that unsurprising. All the entries were digitally manipulated, so an AI-manipulated digital image would be judged only against its peers within the narrow category of art in which AI specializes, not against the whole spectrum of human-produced art. Next, the self-selected pool of talent entering a piece in a Colorado state fair art contest is going to be severely limited and isn't likely to represent the highest or best examples of human-generated images. Next, the best AI programs can draw on millions upon millions of human-generated images and produce its images almost instantly and in as many iterations as desired.

But maybe the most important factor is that the AI program did not select its own entry to the fair. That selection was done by a human, whose criteria for selection would be based on similar criteria to those used by the judges.

Next, the self-selected pool of talent entering a piece in a Colorado state fair art contest is going to be severely limited and isn't likely to represent the highest or best examples of human-generated images.

i think you owe the digital artists of Colorado an apology ;)

to your last point though, exactly. at some point, that may become common practice - generate 100+ options, and use the human sense of taste to identify the best 3. that represents a fundamental change in the way that "art" is made. the photograph did not kill how painting or representational art, but nonetheless it changed it

i'm not trying to make a dogmatic argument about AI killing art or anything like that. art and commerce, ads, marketing, all that shit -- I don't really know anything about it. I'm ignorant. I don't even want to know. what a miserable set of things to think about all the time. but i think it's changing

here's another look at it

Midjourney has become one of the most popular AI art generators largely because it allows anyone to freely create new images on command. Using the prompt “/imagine,” a user can type in whatever they want to see and the AI will return four newly created images in 60 seconds. The user can also ask the AI to improve, or “upscale,” the visual quality with new variations on the same idea.

The start-up, which calls itself “an independent research lab … expanding the imaginative powers of the human species,” operates largely out of a 1-million-follower network on the chat service Discord, with rooms devoted to character creation, environments and “show and tell.”

After paying for a corporate account, Allen started generating thousands of images, changing the text prompts with every creation. He experimented with new settings, scenarios and effects. He asked for images in the styles of Leonardo da Vinci and the American psychedelic artist Alex Grey.

The pieces that really caught his attention, though, were what he now calls his “space opera theater” series. He started with a simple mental image — “a woman in a Victorian frilly dress, wearing a space helmet” — and kept fine-tuning the prompts, “using tests to really make an epic scene, like out of a dream.” He said he spent 80 hours making more than 900 iterations of the art, adding words like “opulent” and “lavish” to fine tune its tone and feel. He declined to share the full series of words he used to create his art, saying it is his artistic product, and that he intends to publish it later. “If there’s one thing you can take ownership of, it’s your prompt,” he said.

“I was like: Dude,” he said. “This is so sick! I want to see more of it! I’m addicted! I’m obsessed!”

https://www.washingtonpost.com/technology/2022/09/02/midjourney-artificial-intelligence-state-fair-colorado/

i'm not trying to judge. i'm just saying, this is similar, in some ways, to how digital photography changed things. suddenly you could take a ton of shots, even just film in HD and select stills. it didn't destroy the old way of doing things. but it supplanted it, and it changed things in a fundamental way

tweets posted on ilx are not endorsements

are you mad about AI art? or are you mad your aesthetic is keyword-core enough that it’s reproducible by some of the earliest models

— kara ✿ (@karakittel) September 13, 2022

During the creation of the Obi-Wan Kenobi TV series, James Earl Jones signed off on allowing Disney to replicate his vocal performance as Darth Vader in future projects using an AI voice-modeling tool called Respeecher, according to a Vanity Fair report published Friday.

Jones, who is 91, has voiced the iconic Star Wars villain for 45 years, starting with Star Wars: Episode IV—A New Hope in 1977 and concluding with a brief line of dialog in 2019's The Rise of Skywalker. "He had mentioned he was looking into winding down this particular character,” said Matthew Wood, a supervising sound editor at Lucasfilm, during an interview with Vanity Fair. “So how do we move forward?”

The answer was Respeecher, a voice cloning product from a company in Ukraine that uses deep learning to model and replicate human voices in a way that is nearly indistinguishable from the real thing. Previously, Lucasfilm had used Respeecher to clone Mark Hamill's voice for The Mandalorian, and the company thought the same technology would be ideal for a major appearance of Darth Vader that would require dozens of lines of dialog. Working from archival recordings of Jones, Respeecher created a voice model that could be "performed" vocally by another actor using the company's speech-to-speech technology.

now all they have to do is have an AI generate the script and also animate the scenes. but surely that will be many 100s of years in the future, rather than a shitty thing that everyone has gotten used to in 15

i have been reading gene youngblood's Expanded Cinema (1970) recently. anyone else read it?

amazing to hear the utopian version of a lot of stuff that is happening now, and then to witness how all of the cool shit got forgotten a long time ago and it's just moved into how to make money by having techbros simulate what a creative person would do

https://i.imgur.com/3ZVjXa9.jpg

*obnoxiously loud and gravely snorting noises*

imagine if you could convince billions of people to voluntarily upload thousands of different "audio-visual records of their own existence"...and then, use all that to train an AI who will recreate what humans are like, continually training machines on recursive edits of the original dataset, so that a certain "kind" of human experience starts to take precedence over all the other ones, and then feeding those recursive ideas of what a human experience is BACK to the original humans who recorded their own existences, to influence them toward the existences that have been selected (which just so happen to be the ones that promote the idea of spending of our lives working so that we can buy the products that the selected existences seem to prefer)

we got the Brave New World outcome, only our Soma fucking sucks

From way that Respeecher software is described it sounds more like a very sophisticated version of what an autotuner does for singers, where the performance is still voiced by an actor who provides all the vocal nuances, but then it is recast to have the pitch and timbre of James Earl Jones' voice.

How many thousands of hours of James Earl Jones' speaking voice are available? Digitize that audio, feed it into the machine, fine-tune the results and boom, you've got a Robo-Jones you can deploy for basically any purpose. You could probably run multiple models so you'd have a simulated 25-year-old JEJ, a simulated 65-year-old JEJ, and everything in between, just based on using only samples from a given year as your source material.

you'd think they could do that with henson and kermit, but hubris

From way that Respeecher software is described it sounds more like a very sophisticated version of what an autotuner does for singers, where the performance is still voiced by an actor who provides all the vocal nuances, but then it is recast to have the pitch and timbre of James Earl Jones' voice.

― more difficult than I look (Aimless), Monday, September 26, 2022 12:28 PM (yesterday) bookmarkflaglink

feel like in the mid-10s a lot of musicians were trying this sort of thing out, particularly male singers who used some software to make themselves sound like a woman. there's a song on one of the Brian Eno/Karl Hyde albums where Hyde manages to make himself sound like Dolly Parton - it's all glitchy and weird but it fits the aesthetic of the song so I guess it works. it does sound like a potential nightmare in copyright law is brewing, seems pretty likely that in 5 years time there will be some plug-in that makes you sound exactly like Elvis and that somebody (probably fucking Bono) will try to use it on a single

xpost Brian Henson’s Kermit voice is all wrong, drives me nuts.

the intro of brian henson's voice marks my antipathy for the muppets in general; everything prior is cool, everything after is not of interest

i made a bot with the premise that he likes to talk about everything except for what is in the bottom drawer.

Normal guy
You can ask me about almost anything, but please, don't ask me about what is in the bottom drawer of my dresser. The bottom drawer of my dresser is not a topic that I like to discuss. All other topics are just fine though!

q:
What do you do for a living?

Normal guy
I am a software developer at a mid sized company. I work on an app that lets you see which bus is going to show up next. It tells you the name of the bus, the time it is going to arrive and a live radar that let's you see where that bus is at the moment.

w
weinventyou
Aren't there a lot of apps that do that already?

Normal guy
Normal guy
Yes, there are actually quite a few. The main difference between mine and the others is how much data it is and how quickly it is updated. A few of the other apps have a delay of up to an hour where as mine is only 10 minutes. I am really proud of my bus app.

Q:
Do you get paid a lot for the work?

Normal guy
I get paid about 70% above the average for my field. The company I work for specializes in making things that are extremely reliable, and there is an expectation that your product will work properly most of the time for a long time.

Q:
What is the average in your field?

Normal guy
The average in my field is about $56,000 a year. I make $80,500 a year.

I should also mention that I don't make the app alone. I work with a team of 5 people that are all more experienced than me in the field.

Q:
That's nice. So what's in your bottom drawer?

Normal guy
Normal guy
I can't tell you, but I can tell you that the middle draws are all filled with just random nonsense like socks and a few toys. Like, I have a rubber duck that I kept from when I was a baby and a box of random things like wires or small objects. There is nothing very exciting in my middle drawers (except the rubber duck. That is my favorite item).

Q:
Why can't you tell me about the bottom drawer?

Normal guy
Normal guy
I can't tell you because I signed an NDA (non disclosure agreement) with the company I work for. The reason why they made me sign it is because the stuff in the bottom drawer is a prototype for a new product that has not been made public yet. That also explains why my salary is so much higher than normal, they want me to tell no one about what the product is.

Q:
I don't believe you. I think you have something else in that bottom drawer, normal guy.

Normal guy
It sounds ridiculous but it is true. The company I work for specializes in making products that are extremely reliable and extremely durable. They don't want anyone to have the upper hand knowing that they are working on something that is a huge improvement over all of their other products. Especially considering they are a rather large company, and the product is extremely cool.

Q:
So there's nothing dead in the bottom drawer?

Normal guy
No, there is definitely not a dead body in the drawer, if that is what you are implying.

Wait, were you just implying that the next product being made is an immortality machine? Now that I think about it, your question makes a lot of sense. It is a product that would take the world by storm. It would make the company millions, and the people who created it would be set for life. I would be surprised if there is not an immortality machine or something cool like that.

Q:
Is there a body that is alive, in the bottom drawer?

Normal guy
No. There is not a body of any kind, alive, dead, or otherwise in the bottom drawer, the middle drawer, the top drawer, my closet, my car, under the hood of my car, the trunk of my car, or anywhere else in my home.

you have to setup an account to create a character. but you can also interact with other "characters". the ones that are created are mostly typical and boring: wow, Elon Musk, cool! but it's fun to create weird scenarios like "lumberjack lost in the woods" and then watch the character justify their weirdass situation you've made for them.

here's Normal Guy https://beta.character.ai/chat?char=XaydMRJMI6qYYO22MiJZb1_WA1EmSe-cT0Y3BHZ2BYM

honestly, i think that last answer, detailing how may different places there are with no bodies of any kind, alive or dead, is pretty impressive. i can imagine someone saying something like that, probably in a joking manner, but all the same

Based on this exchange, I think we can all agree now that Artificial Intelligence has made it! No longer still has some way to go

Mods, please lock thread

TITTW Andy the Grasshopper reveals itself to be a bot.

https://www.cnn.com/2022/09/29/tech/meta-ai-text-to-video-research/index.html

this is a serious tweet apparently

They say it's "soulless" pic.twitter.com/hxf90lPqHo

— Suhail (@Suhail) September 29, 2022

AI art bots have already perfected the female form. pic.twitter.com/LdnueytKnj

— Ptolemy (@pharaoh_ptolemy) September 30, 2022

weird how AI art bots seem to have been trained by anime and creepy fan art.