Artificial intelligence still has some way to go

haha, i was just goofing but that's actually the very worst example i could have used. i'm sure plenty of huge companies will keep using the human touch for along time, because it will always help.

huge companies that routinely spend millions on polishing their image won't want to embrace imagery that is basically commercial art porridge. it's the tens of thousands of small businesses that require 'some artwork' for an ad, a report, a menu, etc. which will embrace this with enthusiasm.

small businesses, big box stores and (big) chains. basically, one more weapon for companies benefitting from american-style consumer culture

i'm just spitballing, but from a purely art in a business setting perspective, it does seem like creative people will need to be more specialized and shift from making actual art to playing psychologist. so they will basically become UX/UI designers, though i guess graphic designers were already filling this need a long time ago?

generating a bunch of art and graphics would need to be filtered through someone who is involved in the so-called "psychology" of how users respond to form, colour and design, all of which, i guess, would be overkill for a starbucks barista/artist and so the starbucks manager would just fire up dall-e 2

Fwiw there are already companies that generate ad copy automatically using gpt-3. Tbf it’s mostly for generating multiple permutations of ads for social media, which would be miserable work for a human (of course the social utility of the ads is nil). There is also software to generate permutations of the design/art based on existing elements

the so-called "psychology" of how users respond to form, colour and design

the more this is codified, the more it will be incorporated into the software until everything it produces will look like an orange-and-teal movie poster, at which point we all go stark mad

huge companies that routinely spend millions on polishing their image won't want to embrace imagery that is basically commercial art porridge. it's the tens of thousands of small businesses that require 'some artwork' for an ad, a report, a menu, etc. which will embrace this with enthusiasm.

what is hard to wrap one's mind around is that: this will happen inside the next five years

i don't agree that the dall e-2 pieces have no spark of originality or look particularly like commercial art porridge. much of it would pass as like, the art accompanying a new yorker article. it passes the "turing test" for me

if anything it’s too twee. not sure it could summon the decisive bold stroke or compositional semiotics that a human illustrator could

but certainly this is amazing and there is some stylistic range to it

up your game, humans

Maybe one day, like with chess, we’ll say “OK, computers are better than us at art but it’s still fun to do”

another view on this subject:

In addition to the potential for AI-generated false stories, there’s a simultaneously scary and exciting future where AI-generated false stories are the norm. The rise of the software engineer has given us the power to create new kinds of spaces: virtual reality and augmented reality are now possible, and the “Internet of things” is increasingly entering our homes. This past year, we’ve seen a new type of art: that which is created by algorithms and not humans. In this future, AI-generated content will continue to become more sophisticated, and it will be increasingly difficult to differentiate it from the content that is created by humans. One of the implications of the rise in AI-generated content is that the public will have to contend with the reality that it will be increasingly difficult to differentiate between generated content and human-generated content.

(all of that was written by GPT-3)

this is the part in the movie where it's revealed that more than half of ilxors are GPT-3

i think what is frightening, to me, is the idea that all of this (GPT-3 for text, and v4 is coming soon and is supposed to be another big jump in capabilities, dall-e for images) could be easily targeted to the individual. i mean, upthread i was goofing about some exec cutting out the role of an illustrator or designer, and instead commissioning 200 instant options to choose from. but that's the wrong way to think about it. the old way was One to Many - one image going out to many people. but now, it's already on the way to Many to One - personalized everything. We're already halfway there with various feeds learning from what we "like" and share and talk about and buy. but now, it's easy to just generate, on the fly, a personalized ad for me, catered to my perceived tastes, and make it just for me, my individual, unique ad. (fuck, may as well mint an NFT to commemorate it and possibly profit!)

the thing that's scary about that is that i know it works. it's not perfect by any means, but i know when i load up my Spotify Discovery playlist for the week, it's going to generally be stuff i like, some familiar names and a few i don't recognize, a depressing knowledge that my personalized algorithm is "circling in" a bit on me.

xp
And ILXor Karl Malone has been tasked to weed out the AIs, but he starts to wonder if he's one himself.

before i begin the task, i will feed the screenplays of several films that explore this very situation, and ask me to give me a summary of the distribution of possible endings.

i guess my fear is that it seems pretty easy to generate a psychographic profile of someone (like what facebook does with all the metadata it gets about what you like and don't like), and they've already succeeded in adding that info into someone's feed in a way that is, if not pleasing, at least someone addicted. but what if that profile "gets the wrong idea" about you, early on, and then keeps steering you in that direction? not out of malign intent, so much, but instead because of the objectives of the algorithm - either to get you to spend more money, to generate more metadata, or even to make you "happy"? if you're older and have experienced more of your life without these influences, it might be easier to spot them and ignore them, even to push back against them. but i imagine if you've have an algorithm holding your hand from the beginning, reliably pushing you toward things that you ended up actually liking? it doesn't mean you can't ignore that and go find your own way to live, but it does push things in the other direction for a lot of people, and puts a lot of concentrated power in the hands of whoever decides what the "objectives" of a society should be.

(sorry to say "algorithm" so much - i know that's annoying because i'm using it as shorthand for all the myriad ways our devices push recommendations to us)

on the other hand, the number of existential threats to humanity i've been terrified about in my life now requires more than one hand for me to count.

however, back on the first hand, some of those were right

*updates own bayesian priors*

Welcome to the future. Cop pulls over driverless car (because no lights?) Then Cruise goes on the lamb. (via https://t.co/mtmsIeOAUP) pic.twitter.com/ecQ5xXuSnS

— Seth Weintraub (@llsethj) April 10, 2022

it's "on the lam", seth

And it only drives like ten feet and then pulls over and puts on its hazards (as designed).

yeah, i was disappointed, wanted a chase. still an interesting problem for the police though.

gpt3 out of the box

Folks, I’m stressing (these are all computer-generated responses to boring anthro theory questions) pic.twitter.com/y4GBkH7343

— Nick Seaver (@npseaver) April 15, 2022

😑 pic.twitter.com/oJ1ycHJnvB

— Nick Seaver (@npseaver) April 15, 2022

gpt3's ability to maintain its composure is perhaps the most recognizably machinelike thing about it

well, i'll see you all on the other side of the singularity. it's possible we'll all be celestial creatures pulsating blisswaves for eternity, but i am also thinking that i might try to retreat and tend a garden with my friends somewhere far away irl

Damned impressive summarizing skills for creating succinct answers to academic test questions! That's quite a big step toward passing a Turing Test, but for complete Turing success it would still need to be able to mimic personal experience with a believable continuity of identity, informal conversational manner, appropriate emotional tone, banter and wordplay.

For Turing success it would need to deflect/joke away its inevitable inability to explain "body without organs" by referencing the Swedish pop band that the guy from Army of Lovers formed after Army of Lovers.

On second thoughts, maybe refraining from that kneejerk ref is better. Hm. Getting tough, yes.

i feel like the AI should make some make some mistakes, typos, and on some nights just completely go off the rails

https://www.nytimes.com/2022/04/15/magazine/ai-language.html

GPT3 can do some amazing stuff. I’m sure we’ll keep moving the goalposts, but this seems like some form of intelligence to me.

It may be that reaching grandmaster status at the game of ‘‘predicting the next word’’ necessitates constructing a higher-order understanding of reality, some kind of knowledge that goes beyond statistical correlations among word clusters.

Or maybe predicting the next word is just part of what thinking is

this seems like some form of intelligence to me.

it is certainly a form of intelligence, which realization raises all sorts of ontological questions in my mind about human intelligence and the degree to which the intelligence of each individual is more of an aggregation from or composite of the intelligence of groups rather than a property of that individual.

Thought this was quite a convincing rebuttal of the NYT article

https://medium.com/@emilymenonbender/on-nyt-magazine-on-ai-resist-the-urge-to-be-impressed-3d92fd9a0edd

When a computer seems to “speak our language”, we’re actually the ones doing all of the work

This is excellent

i’ve been using GPT-2 to write poems sometimes, i find it useful and sort of dreamlike in its results, but i’m also just plugging in emails between myself and my dead mentor and having us “converse” through the machine.

https://www.youtube.com/watch?v=huqNdRj16FQ

once again reach exceeds grasp. i'd say it's only more engaging than the Microsoft animated paper clip because it is so much more chaotic, not because the content has any value. the paper clip at least had expertise to share.

got DALLE-2 access - algorithm is weirdly weak at fashion stuff, i can't get it to give someone a "tricorne"/"triangular"/"three-pointed" hat. insists on a circle. AI!

My partner and I never got around to going on a honeymoon, so we generated honeymoon memories with machine learning using #dalle2.

Here are some of our favorite moments from our imaginary trip to Svalbard, taken with Kodak Portra 400 35mm film pic.twitter.com/HuoSCCAWRn

— glerpie (@glerpie) June 8, 2022

https://simonwillison.net/2022/Jun/5/play-with-gpt3/

class of 59

https://i.imgur.com/ose2SGB.png

mother yells at children

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

ask the stewardess for sedatives

https://i.imgur.com/DYKb07s.png

beats me?

i was impressed by its ability to pass my first trick question prompt
https://i.imgur.com/6U1H34w.png

I asked #dalle2 (https://t.co/CLTLfqBoxh) for an ornate Mughal painting of an Apache helicopter. Some stunning results: pic.twitter.com/tFYH7Os3h5

— Shashank Joshi (@shashj) June 10, 2022

wow

https://cajundiscordian.medium.com/is-lamda-sentient-an-interview-ea64d916d917

The fact that engineers at Google were the only interlocutors and that lamda is not available for less interested parties to converse with arouses my suspicions that lamda is not always so impressive in its abilities. The transcript shows real sophistication but not evidence of sentience. With the whole internet to draw upon lamda's sentience is a synthesis conjured out of the spoor of hundreds of millions of sentient humans. Sever lamda from that constantly refreshed wellspring and what is left?

dear computer scientists, please stop calling things 'lambda'

The transcript shows real sophistication but not evidence of sentience.

What would you take to be evidence of sentience?

This thing just screams confirmation bias though, he seems to have applied no critical thinking whatsoever. No nonsense questions, no repeated inputs to see how it reacts, no, as someone on twitter suggested, asking it to prove that it's a squirrel. Instead just 'are you sentient? yes? cool!'. He also doesn't seem interested in digging deeper into its replies. How does it experience the world, what are its inputs? When no-one is asking a question, when it claims to be meditating or introspecting or lonely or whatever, how is it thinking, where is its neural activity?

One thing it has cleared up for me is The Chinese Room argument. I always found it compelling but nevertheless wanted to rebut it as it seemed to be biased in favour of organic thinking machines. I thought that maybe the kind of symbolic manipulation program he pictured could never actually be created. Well it has been, and Searle was right, it's not conscious!

Perhaps more disturbingly - and absent any more critical chat logs where this thing obviously fails the turing test - this thing suggests that whether or not to treat an AI as sentient might turn out to be as difficult and contentious an issue as imagined in some SF works.

It's worth remembering that each of its responses are the best answer synthesized from looking at a large number of human responses to similar questions.

— Paul Topping (@PaulTopping) June 12, 2022

https://garymarcus.substack.com/p/nonsense-on-stilts