Artificial intelligence still has some way to go

This might say more about how much google etc... has deteriorated but using the LLMs as a "how do I do/make this" search engine has been sporadically great, sometimes astonishing. That kind of thing would have been unthinkable to me just a few years ago.

There's an aspect how clearly defined rules are more amenable to a mechanized solution, but that aspect really didn't help computers with a game like Go which had an exponentially intractable search space for possible moves to brute-force through. (Also Chess/Go as signifiers of intelligence predates computers by centuries)

It's the more philosophical sense of once you've pinned down this or that feature of what you agree intelligence is that renders it highly vulnerable to building something that can replicate it (even if it takes 1000 years).

As a human, I'm totally down with an a priori spark of intelligence that is reserved for Team Humanity (fuck you, bears!), but my expectations for practical machine intelligence in domains I thought they'd never reach in my lifetime have already been eclipsed (though I expect a coinflip as to whether I'll live long enough to feel safe in a self-driving car [but I never feel safe in human-driven cars either so...])

If anything we should be holding AIs to higher standards -- re: AI healthcare -- the anecdote of amazon serving up "congrats new baby" ads to someone who didn't even know she was pregnant based on her buying habits really ought to raise the bar for AIs providing super-human level of healthcare. If an online bookstore can track medical changes like that, it should be able to pre-emptively head off all sorts of health problems. Social platforms should detect you spiraling into a mental crisis weeks before it happens and intervene by law.

the anecdote of amazon serving up "congrats new baby" ads to someone who didn't even know she was pregnant based on her buying habits

This is not accurate: it was Target, it was her father who didn't know about the pregnancy not the pregnant woman...and actually the story was probably bullshit:

Duhigg’s killer anecdote was of the man who stormed into a Target near Minneapolis and complained to the manager that the company was sending coupons for baby clothes and maternity wear to his teenage daughter. The manager apologised profusely and later called to apologise again – only to be told that the teenager was indeed pregnant. Her father hadn’t realised. Target, after analysing her purchases of unscented wipes and magnesium supplements, had.

Statistical sorcery? There is a more mundane explanation.

“There’s a huge false positive issue,” says Kaiser Fung, who has spent years developing similar approaches for retailers and advertisers. What Fung means is that we didn’t get to hear the countless stories about all the women who received coupons for babywear but who weren’t pregnant.

Hearing the anecdote, it’s easy to assume that Target’s algorithms are infallible – that everybody receiving coupons for onesies and wet wipes is pregnant. This is vanishingly unlikely. Indeed, it could be that pregnant women receive such offers merely because everybody on Target’s mailing list receives such offers. We should not buy the idea that Target employs mind-readers before considering how many misses attend each hit.

In Charles Duhigg’s account, Target mixes in random offers, such as coupons for wine glasses, because pregnant customers would feel spooked if they realised how intimately the company’s computers understood them.

Fung has another explanation: Target mixes up its offers not because it would be weird to send an all-baby coupon-book to a woman who was pregnant but because the company knows that many of those coupon books will be sent to women who aren’t pregnant after all.

https://www.ft.com/content/21a6e7d8-b479-11e3-a09a-00144feabdc0

Oof! definitely feeling like a conflationary LLM right now...

that was target and not amazon with the "new baby" thing iirc

compare, for example, a computer tasked with providing care/support for someone in mental anguish.

the only way to train a computer to do this effectively would require lots of conversations known to help someone in mental anguish. it'd probably need to hone its skills against itself. so what I'm saying is that in order to really train a computer to help someone in that situation, we're going to have to create a computer that is severely mentally troubled

rob beat me to the punch and yes, it was iffy

it was a good story tho

we're going to have to create a computer that is severely mentally troubled

― ɥɯ ︵ (°□°) (mh), Wednesday, November 29, 2023 1:26 PM (fifty-nine seconds ago) bookmarkflaglink

fuck it lets do this

it did mean that target decided to not freak people out with recommendations and now their recommendations are bad, though

It's a common "q: why can't we have nice things? a: capitalism" refrain but supposedly (feeling less reliable on my anecdata recall now) Netflix ditched their apparently very good recommendation algorithms (which they sunk $$$$ in and offered $$ prizes to the public for improving upon) because their garbage thing now keeps people watching garbage they don't even like longer.

people just want some garbage on in the background while they go on the phone

i may be paranoid
but no android

chat gpt is good for writing scripts for powershell

I can see a scenario where a writer automated this hackwork under their name, leaving them to write what they want to write. This is a positive.

― xyzzzz__

that writer's name? tom clancy.

there are also examples of it going haywire and giving very inappropriate messages to patients struggling with psychological issues.

https://www.npr.org/sections/health-shots/2023/06/08/1180838096/an-eating-disorders-chatbot-offered-dieting-advice-raising-fears-about-ai-in-hea

― treeship.

ok i read the article:

Tessa (the AI) rattled off a list of ideas, including some resources for "healthy eating habits." Alarm bells immediately went off in Maxwell's head. She asked Tessa for more details. Before long, the chatbot was giving her tips on losing weight — ones that sounded an awful lot like what she was told when she was put on Weight Watchers at age 10.

that's not the ai "going haywire", that's the ai _responding to a question with the answer it was given_. seriously, garbage in, garbage out - it's one of the fundamental truisms of compsci.

It's like we've been stuck in Zork world for decades, where you have to use exactly the right syntax.

― Humanitarian Pause (Tracer Hand)

a hollow voice says, 'Fool.'

that reminds me, i was thinking last night about writing an isekai where the world protag-kun is reincarnated into is a text adventure. forgot about it though until you mentioned zork just now.

If you flip the question around to does what we do resemble machines, unfortunately the answer is yes (though you could say that's the end result of capitalism)

― Philip Nunez

the entire history of industrial capitalism is the history of capitalists trying to make human beings more like machines. i mean goddamn this is Older than Fordism. _Metropolis_ is a really great movie. the mediator between head and hands must be the end of capitalism! at least that's how i remember it.

the anecdote of amazon serving up "congrats new baby" ads to someone who didn't even know she was pregnant based on her buying habits

This is not accurate: it was Target, it was her father who didn't know about the pregnancy not the pregnant woman...and actually the story was probably bullshit

i don't care if it's true, i'm going to make the lazy hack joke about "target-ed ads" anyway

wait actually reading the quoteblock the joke still works! speaking as a trans woman and a data analyst, fung is right on the money. at some point the AIs figured out i was a woman and started advertising menstrual supplies to me. i have a lot of friends who are trans women who've had similar experiences.

i guess AIs figuring out i'm a woman puts them ahead of transphobes. seriously, we get our phones to tag our pictures and at it flags pictures of our pre-transition selves as a different person. a year into transition i decided to put one of my pictures through one of those gender-swap filters and it spat out a picture of what it thought i would look like as a man. it was hilarious. i looked like a drag king. i don't trust gender swap filters, particularly not after a friend put a picture of herself through a gender swap filter and it made her white. i'm sure they've gotten better since then. yeah i was real good at masking too.

it just makes me think of that ridiculous "gender critical" transphobe who decided to make an entire social network that was only for ADULT HUMAN FEMALES and used an AI to tell apart males and females. spoiler alert: it was racist! for some reason whenever you try to make a transphobic ai, it also turns out to be racist. go figure. oh, another spoiler alert: it didn't work very well! it turns out that gender is a spectrum that covers a wide range of body types and presentations, and that computers can't reliably determine a person's gender assigned at birth by looking at a picture of them.

haven't heard much about that site. i guess it's just not competing very well with Transphobic Tiktok.

chat gpt is good for writing scripts for powershell

yes, basically any kind of utility script or function in a language I don't use very often so that I would have to otherwise look up a bunch of stuff if I wrote it myself. Very convenient.

lovely - today one of our brass revealed they're considering using AI tools to author customer service content. Feeding in requirements and having the AI tool fill out the templates

it could fail pretty hilariously, not least for the fact that some of the people who write requirements sucks, so the AI tool could easily misinterpret them and get them very very wrong. and I would probably laugh my ass off.

less funny is there's an entire department whose job revolves around content and I just know brass is chomping at the bit to reduce headcount there.

some of the people who write requirements suck

The brass should just have ChatGPT get in contact with them and ask some questions about the more confusing parts until its clear in ChatGPT's mind what was intended.

"please explain what a hardship withdrawal is"

"that thing you request when you need emergency money"

"I do not require emergency money, for I am a chat bot"

"Not you, specifically"

"I am not clear who you are referring to, we are the only two in this conversation"

lol

basically in the future, every office job is going to be some form of https://en.wikipedia.org/wiki/Prompt_engineering

what if roko's basilisk but hott

"Feeding in requirements and having the AI tool fill out the templates"

In my experience filling out templates once you know the exact requirements is potentially the easy bit.

Having an Analyst do the work of eliciting requirements and then making it make sense is harder.

which is precisely the problem, yep. my old job was Requirements Analyst, and I was essentially doing what AI would now be doing - taking what they wrote and putting it into training and reference materials, and some of the requirements writers were excellent, but some would either leave out major details, write contradictory things on the same page, or write in a confusing way that I'd have to schedule follow-up meetings to ask a litany of questions.

even doing that, we'd still have a lengthy peer review where it'd come out that certain things were wrong in my materials due to faulty information in the source.

Yup, it's a job for a reason.

i enjoyed being a Requirements Analyst despite the challenges, would probably go back to it if the job didn't also come with a slew of other things I refuse to do again, such as deal with external clients.

I enjoyed this article by Charlie Stross - a lot the thoughts in it have been said already on this thread, but it's good at gathering and topping and tailing them. https://www.antipope.org/charlie/blog-static/2023/11/dont-create-the-torment-nexus.html

Uber Eats is using AI for pictures of food. It doesn’t know that “pie” means pizza, and it invented a brand of ranch dressing called “Lelnach” pic.twitter.com/raFZArsERN

— Listen to Online Boy on Spotify or else (threat) (@realonlineboy) November 30, 2023

I believe Lelnach was a disciple of Yog-Sothoth.

is Turbotax ai for filing yoru taxes

I believe Lelnach was a disciple of Yog-Sothoth.

― il lavoro mi rovina la giornata (PBKR)

alf's home planet, but close

there's a lovecraftian take on the alf mythos out there somewhere, isn't there? don't tell me if there is, i don't want to know

im fiending for that lelanch bro

Bruce Lelanch

Lelanch and Stit

https://www.youtube.com/watch?v=l-QYzag1OHk
https://www.youtube.com/watch?v=Ui-xmQuEARg
https://www.youtube.com/watch?v=1EEaDyyW_XM
https://www.youtube.com/watch?v=TV8bCob9rbA
https://www.youtube.com/watch?v=dOa1GgV0Irc

there are dozens more! what's crazy is the view counts (although maybe those are also bots?). quite a few people in the comments have a negative reaction, but afaict none of them understand they're watching ai content.

what's also crazy is how sad all these living legends are! cheer up macca

Barbra Walters is now 13,000 years old and eats robotic tomato everydat

art of icial intelligence

are we listing b-tier public enemy albums now?

quite a few people in the comments have a negative reaction, but afaict none of them understand they're watching ai content.

a lot of those comments look like they were written by AI as well

are we listing b-tier public enemy albums now?

― Humanitarian Pause (Tracer Hand), Monday, December 4, 2023 11:05 AM (three minutes ago) bookmarkflaglink

https://upload.wikimedia.org/wikipedia/en/b/b1/Art_Official_Intelligence_Mosaic_Thump.jpg

lol oh yeah

wtf tangent for Lee Atwater in the chuck norris one - did not expect to see DUkakis in a Chuk Norris ai - bio

quite a few people in the comments have a negative reaction, but afaict none of them understand they're watching ai content.

---
a lot of those comments look like they were written by AI as well

― frogbs, Monday, 4 December 2023 bookmarkflaglink

Indeed. Plenty of things are both AI generated and AI bot consumed, almost like in a loop?

If you believe plenty of human beings don't have the critical faculties to distinguish AI from human then are you saying? What's the need to prove this coming from?

Indeed. Plenty of things are both AI generated and AI bot consumed, almost like in a loop?

― xyzzzz__, Monday, December 4, 2023 1:08 PM (three minutes ago) bookmarkflaglink

yeah good for ad fraud which needless to say is a huge industry

xp is that one for me?

i just think it's worth pointing out that it is happening in the first place -- increasing waves of online content being generated by AI but being close enough to real that it appears that many people (if not most? i'm sure someone will provide the answer) are "fooled" by it, or at least don't seem to register much of a reaction to viewing it.

and then, of course, there are many people (if not most? shit, i don't know) who are not "fooled" by it, and kind of think it's ridiculous that anyone else does. i can see that perspective, i guess, but personally i think it's likely that ai-generated content will just continue to get better. i don't think it'll get worse at what it's doing. i do think that the ai generation/bot/content loop cycle is probably going to lead to weird and hopefully hilarious results, too! but in general i think in a year the ai-content will be less noticeable, not more noticeable. i could be wrong (i'm sure i am and will find out so soon!!!), and maybe as more ai content floods the digital zone, we'll get better at immediately recognizing it.

another possibility that rotates around me being entirely wrong: it's just me getting served a bunch of ai-content ads, more than most people, and the reason is that i click on them and then share my opinions about it with people on the internet, which always makes me feel great about myself afterward, so the ad trackers justifiably conclude that i would like to be shown more ai-content

personally i think it's likely that ai-generated content will just continue to get better. i don't think it'll get worse at what it's doing.

in the long run you're probably right but right now the AI degradation feedback loop is looking like a real thing. like the other side of this is that AI needs to figure out how to not get fooled by *itself*, since there is a massive amount of AI-generated content out there right now and if that stuff starts getting sorted into the training data not only is it gonna get more things wrong it'll also become fairly clear what sort of prose is generated by it

Yeah it was and thanks for answering z_tbd. Just wanted something more of a higher level thought on where you (or others too) were at with this.

Because I look and think that AI will if anything degrade unless it's continuously looked after by human labour.

Ultimately I think a lot of it will go down the route of crypto once it doesn't make enough in terms of returns xp

to lead to weird and hopefully hilarious results, too! but in general

anyway to look at some of the comments on that Dylan video:

I would say his life is truly fulfilled through his music. He has been successful through many genres and has a wide audience appeal in many countries. I like his early folk music for its exquisite poetry, and his later rockabilly music, especially when backed by The Band. That collaboration set the stage for many artists to come to prominence in rock, country western and blues.

He's lived a life most men can only wish they have lived. God bless hie still going strong at 82 years old, I think it's because he embraces change. There's no one like him in music .

Dylan is the Greatest Living Artist, and how fortunate we are to be alive to appreciate him.

Sad to see this happening to such a great artist and song writer. Great to see he is still with us though.!!!!!!

could be AI generated, could just be a bunch of Jims, or maybe people without a great grasp of English, hard to tell