ChatGPT is utterly amazing and scary at the same time

Announcements & Discussions Off-Topic
127

OK, now it is time for something funny (or maybe not).

Over time I am becoming more and more skeptical about AI, especially about what big corporations are cooking for us.

So I made a music video.

This is AI protest song created my me, as an experiment with the AI, what an irony...

Probably the only meaningful way to use AI "art" is to spread awareness of the dangers that AI brings to humans. That was the intention. We don't want our music, art and everything to be stolen shamelesly by AI companies like OpenAI and Suno and then resold to us. Don't let AI corporations to own everything humans created before.

COOKING RECIPE:

STEP 1.

I wrote chorus lyrics myself (that is actual HUMAN creation :-)). The lyrics I wrote were:

We will steal your music,
we will steal your art,
we will steal your job and
will steal your heart

STEP 2.

I prompted ChatGPT:

Write protest song about the world in 2025/2026, about AI stealing everything from every one, "we will steal your music, we will steal your art, we will steal your job and will steal your heart". Social media making society worse. Less real friendship than ever. Less love. Everything is just fake.

STEP 3.

ChatGPT wrote verses. One-shot

STEP 4.

I uploaded lyrics to Suno AI and prompted it as follows:

Polished West Coast rock in the style of Toto: elite studio musicianship, tight and precise rhythm section, funk-influenced bass locked to clean, powerful drums, sophisticated yet accessible harmony with extended chords and occasional key changes, strong melodic hooks, smooth tenor lead vocals with layered harmonies, clean chorus-treated guitars, rich keyboards (Rhodes, analog polysynths, Hammond), mid-tempo confident grooves, high-fidelity glossy 1980s studio production, emotionally restrained and introspective mood, clear arrangement and pristine mix with no lo-fi grit or raw distortion.

STEP 5.

Suno AI generated 4 different songs. I picked ONE of them

STEP 6.

Now the most time consuming part! While the steps 1-5 took just half an hour maybe, this part took more than 24 hours

a) FOR EVERY LINE in the lyrics I prompted Z-Image AI generator (via ComfyUI) to produce image illustrating the line. Each generation took about 45 seconds but it was trial and error as only 1 out of 10 generations was somewhat acceptable.
Overall this step took more than 22 hours

b) I also generated one 5 second AI clip from image generated in previous step (the lady at the end in the park) - that alone took 2 hours

c) I used Davinici Resolve 20 to combine music, all images, video clip to create music video. Especially time consuming part was video-music synchronization, effects and subtitles sync. Video editing and polishing was also about 3-4 hours alone.

14 Likes
128

This the fantastic and so true! The whole world should see this music video and wake up from their digital coma, so life can go back to normal!!!

Jorgen
Singapore

1 Like
129

The implications of using stolen data to train LLMs are way beyond music.

One example that I just thought about:

Open Source GPL-licensed code - this all lands in public source repositories that corporations like OpenAI, Google, Anthropic, Meta, Microsoft use to train their LLM so they can be good at coding and their LLMs land in tools like Copilot or Cursor that some developers use to write their code (according to NVidia CEO, even 90% of NVidia code is now written by AI - which is absurd IMHO).
Anyway, GPL licensed code is provided on the premise that any derivative work should also be GPL licensed and open source.

But WHAT IF (and that is quite realistic scenario), LLM that was trained on GPL code writes code that is based on GPL based solution and such generated code lands (via Copilot/Cursor or whatever) in CLOSED SOURCE software (like NVidia drivers)????

It is illegal to use GPL code in closed source apps as per GPL!

I am not even mentioning the fact that there is a LOT of code in public repositories that are available under STRICT licenses that DO NOT allow use without permission. AI companies do NOT have such permission from authors.
AI training on software subject to restrictive licenses constitutes unauthorized use and breach of contract, regardless of output similarity.

The law must catch up and an OPT-IN rule should be enforced. "No copyrighted or licensed work may be used for training or model development without the explicit, documented consent of the copyright or license holder." All existing LLMs should be banned and permanently deleted. A new training should follow correct guidelines under opt-in rule.

And yet, CEOs of big AI companies are very happy about that, that upto 100% of code is AI generated now in the AI industry

Just in case you have been wondering, AmiBroker source code is 100% FREE from AI-generated code. AmiBroker source code is 100% HUMAN engineered. And quite frankly, I find AI as distraction, as coding requires laser-sharp focus. Also, AI, as GPS in the case of car driving, makes your brain lazy. You certainly don't want that.

I do provide AI AFL Assistant for YOU to use (for those struggling with AFL), but I don't use AI to write AmiBroker's own code. I use AI for the following tasks (not coding related): spell-checking, grammar fixes, style correction, formatting in the documentation and preparing support responses for repetitive common and basic tasks questions (such as "lost key", "how do I reinstall"), but even when they are written with the help of AI, they are manually reviewed and fixed if necessary for correctness. As far as coding goes, I used AI to experiment with languages/platforms I don't know myself such as Flutter, for internet radio hobby app that my wife wanted and I did not have time to write :slight_smile: So as far as education is considered, AI can be useful.

4 Likes
130

image
image472×592 104 KB

131

Yes I have heard about them in November, but I don't know capacity of production and whenever they would sell globally or only to China.

132

Here is another example how those folks at Anthropic (those who claim 100% of their code is AI written) are complete lunatics not knowing anything about proper engineering:
They wrote a TERMINAL (console) application (Claude Code) in REACT (sic!).

For full story why it is such absurd idea see:

https://www.reddit.com/r/BetterOffline/comments/1qpzh9p/claude_code_is_written_in_react/

Second thing, there were studies about actual impact of using AI by experienced developers (5 years of experience on average in the study), and in reality AI made them 19% slower:

image
image2562×1540 333 KB

Source: Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity - METR

More interesting comments with regards how AI "helps" in practice of experienced devs:
https://www.reddit.com/r/ExperiencedDevs/comments/1qqy2ro/anthropic_ai_assisted_coding_doesnt_show/

2 Likes
133

There are two more videos that dropped recently on the subject of dare I say a fraud scheme that is called "AI music":

Youtube: The truth about AI music by Sky News

Quote from the video:

"Fraudsters are using AI to mass-produce tracks and then deploying armies of bots to stream them over and over again, a system where robots listen to robot music, all to generate royalties."

Based on analysis made by Deezer shown in the video upto 70% of streaming of AI music is generated by robots. Just to dilute royalties paid to everyone else.

Also another good analysis, how sick this situation is:

Youtube: Suno, AI Music, and the Bad Future by Adam Neely

The "music industry" has massive plans for growth:

https://www.goldmansachs.com/insights/articles/global-music-revenues-are-forecast-to-double-to-200-million-in-2035

on what basis? Would they promote real artists or would they just sell AI-generated stuff based on stolen music, so they don't need to pay royalties for. The answer seems obvious.

Models based on stolen work should be banned and deleted, if copyright means anything.

Also, did you hear about silent protest album (by 1000 famous artists including Kate Bush, Annie Lennox, Billy Ocean, The Clash, Jamiroquai, Tori Amos, Hans Zimmer ): https://www.isthiswhatwewant.com/

Also, Native Instruments filled for insolvency Native Instruments GmbH is in preliminary insolvency - CDM Create Digital Music
- no suprise here - who would need plugins and music software when AI music is flooding the world. Interesting fact that this insolvency is making waves among music studios that largely depend on their (subscription-based) products. Here we are in subscription-based economy.

4 Likes
134

image
image585×725 79.2 KB

https://x.com/AlexanderTw33ts/status/2018436050935292276
135

Arthur C Clarke predicted it in 1957 short story https://www.dna.caltech.edu/~winfree/ArtisticActivities/Arthur-C-Clarke-1957-Ultimate-Melody.pdf

136

Sam Altman just admitted that "they screwed up" and their recent model was worse than previous Sam Altman Says Oops, They Accidentally Made the New Version of ChatGPT Worse Than the Previous One

137

So it looks now that Bitcoin price is lower than mining cost. The end of crypto?

Maybe that is connected with the fact that huge AI spending is sucking money from every place

So now cryptos are getting converted to 'real money' to fund AI?

4 Likes
138

Back in the day, you would trade the same thing applying technicals. Buy when cheap and sell when overvalued.

In modern era, a tulip is invented and it is pumped, then dumped. Those who enter after the pop will be left holding the bag for decades.

Add to that, many things like Bitcoin in Crypto were designed by limited supply, and to a certain extent includes precious metals.

What the smart-guy on wall street does is to create infinite supply with derivatives.
So the spot no longer controls the price, the derivatives, ETFs, leveraged ETFs, swaps and so many other names(instruments) that even i dont know some.

1 Like
139

Some people still don't believe that the AI would take their jobs.
It is quite amazing that 2500 line of markdown (just a plain text file) can cause 20% drop in Thomson Reuters stock price.

I find it funny that those people did not realize earlier that AI is first to take on lawyers because what they do is just language analysis. That is what LLMs do too.

Those people apparently live under some rock, if they don't realize that what Anthropic has shown was there for at least 2 years. The drop is rational, I just wonder why it did not happen way earlier, as the problem is not 2500 line prompt. The problem is that their job is essentially what LLMs do.

The prompt was apparently this:

1 Like