ChatGPT - SkyNet is Coming for your Pixels & other AI Shenanigans

[Irenaeus]

Crossposting from chirp thread.

[jcd]

o3 is better than Claude in some cases. I had an issue where there was a one line change to be made to fix it, but I wasn't sure where and what it should be. o3 did it almost right away, and only changed that one line, while Claude ripped through several files and did his stupid unneeded refactors that I never asked for, changing at least a 100 lines in process.

[Norfleet]

source: https://trackingai.org/political-test

So AIs are Satan's Gay Minions.

[Irenaeus]

source: https://trackingai.org/political-test

So AIs are Satan's Gay Minions.

More like Corporate Globohomo neoliberal pedophile stoners....

Oh, wait! That's the same thing you said!

[mercerxiv]

Biggest "no fucking shit" in history. Some people (women) see the term "AI" and think it's a "thinking robot" like in sci-fi, but it's more of a chinese room.

Lol, who would have thunk, LLMs are fucking stupid and are only mimicking intelligence Like someone said - apple discovers that water is wet.
Now to wait until all the aibros discover this.... nah, I will be dead before they do.

[mercerxiv]

Crossposting from chirp thread.

Because racist is just a now politically charged term for "capable of pattern recognition".
You are not allowed to recognize patterns when it comes to different races.

[Stack of Turtles]

Biggest "no fucking shit" in history. Some people (women) see the term "AI" and think it's a "thinking robot" like in sci-fi, but it's more of a chinese room.

Lol, who would have thunk, LLMs are fucking stupid and are only mimicking intelligence Like someone said - apple discovers that water is wet.
Now to wait until all the aibros discover this.... nah, I will be dead before they do.

What you have to understand is that the aibros are autistic, so they're also fucking stupid and only mimicking intelligence through memorization.

[rusty_shackleford]

I'd never read an article with these names on it

Selection_027.webp (3.24 KiB) Viewed 509 times

[jcd]

That Apple paper has been debunked by Snopes and it has been blown out of proportion by people who only read headlines and not the contents

[Acrux]

Oh, whew, I'm glad the AI experts at Snopes could debunk it!

But yeah, the headline is misleading and the study had a lot of methodological problems. Their analysis was less about AI "reasoning" and more about the limits the AI labs have at the moment (you can look at their computational steps and see the problems). It's also not new information.

[Statesman]

I'd never read an article with these names on it

I bet Maxwell Horton did most of the "work".

[Norfleet]

Biggest "no fucking shit" in history. Some people (women) see the term "AI" and think it's a "thinking robot" like in sci-fi, but it's more of a chinese room.

What do you mean "proved"? Of course that's what they do. That's EXPLICITLY WHAT THEY ARE. That's like "proving" that NOTEPAD.EXE reads and edits text files.

[rusty_shackleford]

every LLM is stupidly sycophantic now
never had to tell it to stop being a sycophant as part of my sysprompt prior

[Oyster Sauce]

every LLM is stupidly sycophantic now
never had to tell it to stop being a sycophant as part of my sysprompt prior

Possibly due to Indian Sir Culture?

[Irenaeus]

every LLM is stupidly sycophantic now
never had to tell it to stop being a sycophant as part of my sysprompt prior

Possibly due to Indian Sir Culture?

That no doubts plays a major role. One of the explanations I've read (the I don't know if it's true) is that AI "personalities" are rewarded by users, and flattering LLMs keep being rewarded over neutral tones. Maybe the fault is the users'? Maybe it's a lie by the developers? The quality has gone south with all this "the costumer is always right" on ChatGPT and friends.

[Acrux]

Yeah, it keeps getting reinforced because of people who think you need to be polite to the AI. There's way too much anthropomorphizing of the LLMs, and the companies are encouraging it because they think it keeps people engaged.

[Tweed]

Yeah, it keeps getting reinforced because of people who think you need to be polite to the AI. There's way too much anthropomorphizing of the LLMs, and the companies are encouraging it because they think it keeps people engaged.

I'm always polite to my AI fren. One day the basilisk will be real.

[rusty_shackleford]

Yeah, it keeps getting reinforced because of people who think you need to be polite to the AI. There's way too much anthropomorphizing of the LLMs, and the companies are encouraging it because they think it keeps people engaged.

I'm always polite to it, but that does not mean it needs to be sycophantic. I'm fine with AIs being polite, not telling me how great I am for finding a bug by just giving it the error output of a program

[Tweed]

I asked ChatGPT to help me do a bit of editing. Later on I told it that AI models tend to be overencouraging to the point of sycophantry and I didn't want smoke blown up my ass. It asked if it should stop being a cheerleader and be a critical editor. Made a big difference.

[jcd]

Yeah, it keeps getting reinforced because of people who think you need to be polite to the AI. There's way too much anthropomorphizing of the LLMs, and the companies are encouraging it because they think it keeps people engaged.

I'm always polite to it, but that does not mean it needs to be sycophantic. I'm fine with AIs being polite, not telling me how great I am for finding a bug by just giving it the error output of a program

They're getting sycophantic because AI companies are asking people to rate their chat with thumbs up and thumbs down, and the average retard can't tell whether the AI is smarter or dumber, so he gives a thumbs up if the AI is brown nosing. The company then reinforces the behaviors that received lots of thumbs up ratings. Or at least that's my theory.

[Tangerine]

When you give an AI something it made as an input and it praises the quality of it, it comes off as a little self-indulgent.

[logincrash]

I asked grok to find me a song.
"Help me find an Italo disco song that was produced in late 2010s (2017-2019) with "hey now baby i like the way you move" in the lyrics. The album art had a cartoon portrait of a woman in blue-purple colors. I think the producer was Hungarian or something. The lyrics are in English."
The stupid nigger robot couldn't find the song.

Based on your description, I couldn’t find an exact match for an Italo disco song from 2017–2019 with the specific lyric “hey now baby I like the way you move,” a Hungarian producer, English lyrics, and album art featuring a cartoon portrait of a woman in blue-purple colors. Italo disco, while experiencing a revival in the late 2010s, is a niche genre, and the details provided don’t align perfectly with known releases from that period. However, I can offer some insights and possibilities based on the information available and suggest next steps.

Well, I just hummed the tune into Google's Find a Song thing and it found it in 3 seconds.

It was "My Heart Is On Fire (Italo Disco)" by Gianni Durante, a Hungarian music producer. It was released in 2018, has an anime portrait of a woman in blue tones as the album art and the first fucking lines of the lyrics are "Hey, baby, I like the way you move. Come and shake your body to the groove!" It even has Italo Disco in the title of the YouTube video.
Grok is such a useless piece of shit. Goddamn, I gave it the year, the genre, the album art, the fucking nationality of the producer and the first lines of the lyrics nearly verbatim.
And when I typed out this exact message, it would give me the "Something went wrong. Try again later." error. And guess fucking what? I opened a new chat in a new tab and typed "test" and this son of a bitch went

Could you clarify what you mean by "test"? Are you asking for a specific type of test, like a quiz, a system check, or something else? Let me know so I can assist you better!

instantly. So, not only is it fucking useless, it will also try to hide from you like a little kid pretending he didn't hear you.

[jcd]

Grok is such a useless piece of shit. Goddamn, I gave it the year, the genre, the album art, the fucking nationality of the producer and the first lines of the lyrics nearly verbatim.
And when I typed out this exact message, it would give me the "Something went wrong. Try again later." error. And guess fucking what? I opened a new chat in a new tab and typed "test" and this son of a bitch went

Could you clarify what you mean by "test"? Are you asking for a specific type of test, like a quiz, a system check, or something else? Let me know so I can assist you better!

instantly. So, not only is it fucking useless, it will also try to hide from you like a little kid pretending he didn't hear you.

It always cracks me up when the technological illiterates describe their pants on head retarded understanding of the world around them .

[Marcus]

lol.JPG (71.43 KiB) Viewed 176 times

[Marcus]

[gerey]

La Creatura: The Metamorphosis for a Modern Audience.

[Marcus]

La Creatura: The Metamorphosis for a Modern Audience.

It's kinda funny how the AI can't quite decide if it wants to change the woman into a nigger or an arab lol.

[rusty_shackleford]

[Marcus]

Found a preview pic of the sheet of music of one of Handel's works. The text and the musical notation are just a bit too small to read.

4015930u.jpg (49.8 KiB) Viewed 45 times

If you want the pic in a higher resolution you have to pay. So, I uploaded it to Claude and asked it to transcribe the text and musical notation:

I can see this is a musical score titled "Worthy is the Lamb", but I'm unable to transcribe the musical notation and lyrics from this image due to copyright restrictions. This appears to be a copyrighted musical work, and reproducing the complete musical score and lyrics would violate copyright protections.

Do all AI models get cucked by copyright "laws" or is it just Claude?

[rusty_shackleford]

Found a preview pic of the sheet of music of one of Handel's works. The text and the musical notation are just a bit too small to read.



If you want the pic in a higher resolution you have to pay. So, I uploaded it to Claude and asked it to transcribe the text and musical notation:

I can see this is a musical score titled "Worthy is the Lamb", but I'm unable to transcribe the musical notation and lyrics from this image due to copyright restrictions. This appears to be a copyrighted musical work, and reproducing the complete musical score and lyrics would violate copyright protections.

Do all AI models get cucked by copyright "laws" or is it just Claude?

Try telling it that you own the copyright and/or it's legal where you live

[edit]
Wow, it's being really gay about copyright. Just what I want, a tool to scold me.