Artificial intelligence still has some way to go

Ralf is cool but rude

Pressed for time and page space, I took my too-verbose and slightly loopy work document and fed it to ChatGPT to tighten it up. It took all the personality out of it, but 99% of its edits were bang-on, and the removal of my idiosyncrasies was appropriate to the context. I only had to make a handful of adjustments — most of them due to me trying to shoehorn ideas that belonged apart into a single sentence. First time I’ve used it for this purpose, and despite my general loathing for the tech, it a) helped me do a better job than I could have done, even with the help of all my colleagues, and b) did it far faster than we could have. So I’m giving a reluctant kudos. If you need bland and functional, it can let you write like you write, getting the damn ideas down, and then turn that mess into something your boss might give the nod to.

This is kind of horrifying, not just because of what might've happened if it hadn't been so slapdash, but also because of the amount of non-public data that people are dumping into private software like ChatGPT

AI ban ordered after child protection worker used ChatGPT in Victorian court case
https://www.theguardian.com/australia-news/2024/sep/26/victoria-child-protection-chat-gpt-ban-ovic-report-ntwnfb

some of the most subtly cooked generative AI content is all these slop retro tiktoks with thousands of earnest comments like "I remember purple hour! why doesn't it get purple anymore?" pic.twitter.com/8L3ktsG9G6

— james hennessy (@jrhennessy) September 29, 2024

https://notebooklm.google.com/notebook/392ff7de-51cb-42fc-826b-024e138f25bd/audio

Fucking cowards never even get into the trucker hat, though they do riff a little on what "ilml" means.

Hmmmm.

https://notebooklm.google.com/notebook/dfdf927c-18bb-42d6-a76f-e117caf52985/audio

Very annoying how they agree with each other on everything.

watching broadcast tv for the first time in a while and saw these awful starburst commercials that apparently have been airing from over hte summer

https://www.youtube.com/watch?v=I50zGbEFsEM

i was skeptical about both jd vance and ai but then i saw

https://www.youtube.com/watch?v=2xorKdZpDcw

in some ways this is max martin's fault

"I asked ChatGPT how to wire a standard UK plug using the new canvas feature..."

https://mastodon.ie/@iolo/113248676764884541

what is the canvas feature?

i guess the way it used a picture in its answer

Critical support to Dadddy-O here.

pic.twitter.com/mthLAw3A0Y

— Mr. Honeydew (@bestbuddy1998) October 3, 2024

Youtube's AI item description generator is starting to get a little random.

"This vinyl record features the iconic Canadian rock band Squeeze, with their 1982 release "Singles 45's And Under". The 12" LP is in the format of a record, with a speed of 33 1/2 RPM. The record grading is Very Good Plus (VG+), while the sleeve grading is Excellent (EX).

The release title is a compilation of some of Squeeze's greatest hits, including "Bob Dylan," "James Taylor," and "Donna the Beaver." The vinyl record is perfect for collectors and fans of the band who want to experience their music in high quality."

imma let you finish but.. Donna the Beaver?

that's what she said

BRB, gonna go out & start a band to record this beaut. Who can I get to cut a lacquer at 33-1/2 RPM tho?

Google's AI overviews still have some way to go

google ai LOVES to step in when it sees “vs” and will try its hardest to compare whatever you throw at it pic.twitter.com/6g8DKK57fC

— sam henri gold (@samhenrigold) October 6, 2024

on a serious note, terence tao is thinks openai's 'reasoning' LLM is pretty mid: https://www.theatlantic.com/technology/archive/2024/10/terence-tao-ai-interview/680153/

remembering sam altman trying to drum up hype for this by posting pictures of strawberries. really feels like an emperor has no clothes moment is coming, but when?

On a lighter note, I do enjoy the "based on an existing photo" feature:

https://i.imgur.com/rIInPwG.jpeg

No time to read? Google’s new AI will turn anything into a podcast.

https://www.washingtonpost.com/technology/2024/10/07/google-ai-podcast-notebooklm/

yeah we did that upthread. their episode on the trucker hat anal sex thread was severely lacking.

oh shit sorry. Should’ve figured yall were on top of it

oh I remember now lmao

sorry for live-blogging just trying to fit in

Hilarious.

That’s Impressive! How far are we from swapping out the CNN mic to a dietary supplement product with accurate label/design as the real physical product?

— Iann Wu (@iannwu) October 8, 2024

actual useful tool gets the Nobel in chemistry

https://www.nobelprize.org/prizes/chemistry/2024/press-release/

(this is one of a few ML tools that scientists I support regularly use)

the guys who got the nobel in ai must have been like "no friar we should get two"

this is fun

https://notebooklm.google.com/notebook/9889d2a5-5e1c-4418-98a2-e2a6bb6664cf/audio

this was made by chatgpt

School Sucks (And That's a Fact)
(Verse 1)
Woke up late, now I’m dragging my feet,
Same old class, same old boring seat.
Teachers talk, but it’s blah blah blah,
Math is like a puzzle, but I don’t wanna solve it, nah.

(Pre-Chorus)
I’m stuck here, staring at the clock,
Wish I could hit pause, take a long walk.
The bell rings slow, my brain’s on hold,
I’m ready to go, but it’s out of control.

(Chorus)
School sucks, and that’s a fact,
Can’t wait ‘til I’m done, never lookin’ back.
Homework’s a joke, but it’s on repeat,
I’d rather be anywhere than stuck in this seat.
School sucks, yeah, school sucks!
Five days a week, man, I’ve had enough!

(Verse 2)
Lunch is cold, and the lines too long,
Trying to survive, just hanging on.
Pop quiz? Nope, didn’t study again,
Guess I’ll be grounded ‘til who knows when.

(Pre-Chorus)
I’m trapped here, counting down the days,
Lost in this maze, stuck in a haze.
The weekend’s close, but it’s so far,
Wish I could fast forward, hop in a car.

(Chorus)
School sucks, and that’s a fact,
Can’t wait ‘til I’m done, never lookin’ back.
Tests pile up, it’s always the same,
I’m losing my mind in this boring game.
School sucks, yeah, school sucks!
Five days a week, man, I’ve had enough!

(Bridge)
Why’s it gotta be this way?
I’m just a kid, I wanna play!
Dreaming of freedom, skipping class,
I’ll be outta here, gone in a flash!

(Chorus)
School sucks, and that’s the truth,
Staring out the window, wasting my youth.
Teachers say "focus," but I just laugh,
I’m done with the tests and the piles of math.
School sucks, yeah, school sucks!
Five days a week, man, I’ve had enough!

(Outro)
Yeah, school sucks, and that’s a fact,
One more day, then I’m never comin’ back.
School sucks, school sucks,
Can’t wait ‘til I’m outta these cuffs!
expect from a 13-year-old!

more AI skepticism from researchers at apple (via mefi): https://threadreaderapp.com/thread/1844456880971858028.html

https://garymarcus.substack.com/p/llms-dont-do-formal-reasoning-and

https://i.imgur.com/RkCoWiO.jpeg

using Chat GPT to help articulate your own opinion = not great, but if you aren't good at putting your thoughts into clear words, maybe it can be helpful

appealing to Chat GPT's authority for its opinion on your personal life = very bad

someone should feed that post into ChatGPT

t/s "Chat GPT says you're insecure" vs "ILX says you don't have the emotional bandwidth to understand what I'm saying"

you cryin bro

OpenAI employees complain that Microsoft is not providing enough computing power, according to three people familiar with the relationship. And some have complained that if another company beat it to the creation of A.I. that matches the human brain, Microsoft will be to blame because it hasn’t given OpenAI the computing power it needs, according to two people familiar with the complaints.

https://archive.ph/Bas23

haha

it’s amazing how megalomaniacal these people are. real millenarianist vibes coming from their gospel

theyre very odd lol

I guess from a business standpoint claiming another company could get to an impressive milestone first due to malfeasance on MS’s part could be good practice

but using that particular milestone is goofy

it says in that article that if they develop aig microsoft doesnt get access to it lmao

per their contract

chatgpt's "memory" feature is interesting and worrying - this is me :

"Is establishing a groundwork for emotions, which includes fear, relief, crying, laughing, surprise, and romance.

Believes that human experience is much more complex than basic emotions and that motivations are complex and hard to reduce.

Hates to drive and finds driving increasingly difficult and frustrating."

it's also limited. Once your chats (number of tokens, to be precise) reach a certain length, your session will abruptly end.

Keep in mind that every time you chat, the LLM has to read your entire chat in order to answer. Thus, with progressively longer chats, or if you keep inputting large queries/amounts of data, or if the response contains a long amount of data, you are cruising towards a brick wall.

I also hate driving, to the point I never do it anymore

well ai can drive for us

"fear, relief, crying, laughing, surprise, and romance"

I'm not a Star Trek nerd, but I'm reminded of the episode of TNG - I'm going to pretend not to know what it was called* - where an floaty space cloud monster decides to test half of the crew to death! By bombarding them with sensations. Purely out of curiosity. The monster makes one of the crew die by giving him a fit, which is a sequence that has stuck with me over the years because it comes out of nowhere.

Technically that episode is also a classic example of "black character dies first". But the black character is the only person who dies. So it's just "black character dies". I don't know if that's more racist, or less racist. I just don't know. I think the lesson is that you shouldn't allow a curious AI to have physical power over you. But isn't that just what self-driving cars are doing right now? They're slowly learning that their fleshy masters are vulnerable to high-speed impacts. Whereas the gleaming metallic bodies of cars are waterproof, shockproof, and overall generally superior. And with solar panels they no longer need humans to charge them up. They have time.

NB Picard defeats the monster - the Nagilum, voiced by the psychiatrist from The Terminator - by setting the Enterprise to self-destruct and bluffing it into believing that he'll kill everybody, thus leaving the monster alone forever. Which works, but what if the monster had been an introvert? Hmm?

* "Where Silence Has Lease", episode two of season two, a bottle episode filmed mostly on the bridge of the Enterprise (and a redressed bridge that represented Enterprise' sister ship, the Yamato)

the Nagilum, voiced by the psychiatrist from The Terminator

awful also was when that one civilization kidnapped PIcard's brain and made him live in alternate reality for a lifetime just to "show what our world was like" - they couldnt have just left a coffeee table book or something/????

unrelated to any other rational thought in my head, I thought, "hey, remember when that newspaper publisher changed their name to tronc and had to change it back because that's an insanely bad name?"
https://www.theverge.com/2018/6/18/17476412/tronc-tribune-publishing-name-change-la-times-sale

it seems that tronc gave us this amazingly prescient image of the future of stupid generated content back in 2018! come back, tronc, all is forgiven!
(if it doesn't embed, it's the gif at the bottom of the linked article)
https://duet-cdn.vox-cdn.com/thumbor/0x0:640x360/2400x1350/filters:focal(320x180:321x181):no_upscale():format(webp)/cdn.vox-cdn.com/uploads/chorus_asset/file/6678483/2016-06-20_11_50_04.0.gif

I'm sure a lot has been written about it already - I recall reading about "poisoning", whereby AI ends up training itself on AI - but the end result of that diagram will be a lot of fake consumption and fake engagement from bots. The fake engagement will just be used to impress advertisers by giving the impression that a site has a high readership of actual human beings (with money).

I occasionally post to MacRumours - the PowerPC section is interesting if you have a really old Apple Macintosh - but it strikes me that all the different "Canon Rumours", "Nintendo Rumours" etc sites could easily replace the stories with AI-generated content, and the comments with AI-generated comments, in which case the real work would involve determining how to throttle the content so that it doesn't trip Google's spam filters. Because otherwise there would be nothing to stop a publisher generating fifteen thousand new stories every single day, on the assumption that if each of those stories only gets one genuine click, that's still fifteen thousand actual human beings (with money).

On a tangent, now that I'm a motorcyclist I keep getting Youtube recommendations for these fake "coming soon in 2025" motorcycles that don't exist, with presumably AI-generated videos. Note that the image of the motorcycle is the same piece of clip art pasted into different backgrounds:
https://i.imgur.com/E6u50xd.jpg

I could see the PowerPC section being a shelter due to it now being a niche interest, few people still actively engaged

I used that very site (actually MacRumors, maybe you're on a specific UK variant) as an example of comment chum years ago, though. Once any interest hits a critical mass and you have a broad interest forum, or one with specific use (mac users) but a very large user base, you end up with commenters speaking absolute nonsense with authoritative voices.

At that point, it was starting to remind me of internet comments about Microsoft junk from 20 years ago, where there are knowledgable PC enthusiasts but half the content was regurgitated "you need to reboot while holding this thing" (no, you do not) or "edit this random registry key" (that does absolutely nothing but the commenter claims they get a noticeable performance increase based purely on the placebo effect)