ChatGPT

[Jopsy Pendragon]

You need to ask Suno to make "Music to break phonographs by".

Gave it "Music to break phonographs by" exactly.

It clearly missed your intent and instead exploded the prompt into: "Industrial noise rock with stomping mid-tempo groove, warped tape-slam percussion, detuned bass drones, and fractured guitar sheets that lurch from verse to verse; verses grind with sparse machine hits, the pre-drop narrows to feedback hiss and sub rattle, then the main section slams in with distorted drum loops and brutal unison accents, Ragged shouted doubles, chopped gang replies, and occasional pitched-down shouts, Ear candy: reverse cymbal swells, vinyl crackle bursts, metal scrape transitions, Gritty, blown-out, ultra-compressed mix with a dirty basement edge"

Weird sorta hybrid of "Art of Noise" and "Skrillex". It kinda works for me. I have no class.

Spoiler

Needle Snaps

 

[Bartholomew Gallacher]

I tried "Make music to make Trump's brain explode" as prompt in SUNO. Amazingly it generated a song including lyrics. Here it is.

Style prompt turned into: Hip-hop diss track with a menacing boom-bap swing, dusty kicks, snapped snares, and a tense bassline; verse rides sparse drums and chopped crowd-noise textures, pre-chorus strips to handclaps and a filtered piano sting, chorus hits with stacked gang shouts and a blunt repeatable hook, Lead vocal is dry and close-mic in verses, doubled and snarling on the hook, with ad-lib pings and short delay throws; risers, tape stops, and reverse cymbal swells bridge into each drop, Mix is gritty, punchy, and in-your-face

Mar A Lago Mirage

Listen and make your own on Suno.

suno.com

 

[Free]

So this seems to be good news.

ChatGPT starts blocking direct requests to copy an author's style

New behavior capturing a writer's "broad qualities" could have legal implications.

arstechnica.com

But the internet - well, certain mostly idiotic parts of it - doesn't like it!

Hack "Writers" Fuming After ChatGPT Starts Refusing Prompts to Copy a Specific Author's Style

AI "writers" are seething after ChatGPT began refusing requests to mimic a specific author's style, living or dead.

futurism.com

 

[Jolene Benoir]

Elon Musk sues Minnesota over law banning AI ‘nudification’

Trillionaire Elon Musk is suing Minnesota over a new law that bans AI “nudification” technology and is set to go into effect Saturday, August 1.

couriermn.com

In the last year alone, the Internet Watch Foundation found an 26,362% increase in the spread of child sexual abuse material (CSAM) online, and Elon Musk’s Grok AI is a prolific suspect in that trend.

Fuck this monster. He wants his AI to continue to be able to take the clothes off women against their will AND chidren.

 

[Innula Zenovka]

I'm increasingly impressed with ChatGPT 5.6 Sol, which I use to help me develop complicated LSL projects. I recently introduced it to the newish function llRezObjectWithParams, which I told it to use in a particular script in preference to the older llRezAtRoot(), so that I could take advantage of REZ_PARAM_STRING to pass string parameters directly rather than have to mess about opening chat channels and listeners.

It admitted it hadn't heard of llRezObjectWithParams before, spent a minute or two digesting the wiki article, and then came back full of enthusiasm for all the cool things it could do with the new function. And since then it's come up, completely unprompted, with some very useful and imaginative solutions to problems by using the function, along with REZ_PARAM_STRING, in ways I'd not thought of to solve otherwise tricky problems. Whatever this is, I think it's more than the stochastic parrot characterisation (or maybe it is, but in that case I don't think "stochastic parrot" is a helpful way of characterising it).

 

[Jopsy Pendragon]

I'm increasingly impressed with ChatGPT 5.6 Sol, which I use to help me develop complicated LSL projects. I recently introduced it to the newish function llRezObjectWithParams, which I told it to use in a particular script in preference to the older llRezAtRoot(), so that I could take advantage of REZ_PARAM_STRING to pass string parameters directly rather than have to mess about opening chat channels and listeners.

It admitted it hadn't heard of llRezObjectWithParams before, spent a minute or two digesting the wiki article, and then came back full of enthusiasm for all the cool things it could do with the new function. And since then it's come up, completely unprompted, with some very useful and imaginative solutions to problems by using the function, along with REZ_PARAM_STRING, in ways I'd not thought of to solve otherwise tricky problems. Whatever this is, I think it's more than the stochastic parrot characterisation (or maybe it is, but in that case I don't think "stochastic parrot" is a helpful way of characterising it).

I'll have to give it another try. Last effort, it simply could not formulate a functional llParticleSystem() parameter list no matter how much hand holding I gave it.

 

[Innula Zenovka]

I'll have to give it another try. Last effort, it simply could not formulate a functional llParticleSystem() parameter list no matter how much hand holding I gave it.

Worth a try -- it's improving all the time. This time last year I was using ChatGPT 5.1 or 2 on a project that I had to drop because both it and I got stuck on something, and other things needed attention. I returned to the project after either 5.3 or 5.4 and it immediately spotted the issue and fixed it for me.

More recently, 5.6 Sol is spookily good -- not only is it coming up with imaginative solutions using new LSL functions I've told it about, but it's also a huge help in debugging stuff. Some of it is just that it's very good at spotting not just those irritating typos that take ages to find (e.g. typing "x = y" when you mean "x == y" and vice versa) but it also finds more complex issues like potential race conditions.

That's saved me hours this weekend. I've just given it the code, told it what's supposed to happen and what is actually happening (or not) and it's not only given me an explanation of where it thinks the race condition is, and why, and detailed patches, but also suggested some debug lines to add to the code before patching it, and told me what to look for to prove it is the race condition it suspects. Right every time so far.

 

[Jopsy Pendragon]

Unrelated, I've been giving some random thought to a rating system, for ai use, at least in as much that it applies to writing. There seems to be a binary classification out there and a rudimentary scale, but here's what I've come up with so far.

AI-0: "minimal" no computer assistance beyond text entry, storage, transmission, spellcheck and formatting. Some online dictionary/thesaurus/search engine use.
AI-1: "technical" grammar/syntax/style checked by programs that may or may not rely on AI. "Light" copy editing
AI-2: "advisory" ai used to provide the author with editorial feedback on tone, style, clarity, logic, and rudimentary fact checking, but no generative content added to the work.
AI-3: "assisted" ai used to rewrite/reword the author's original text in all or part of the work, without changing the underlying ideas opinions claims concepts or concerns.
AI-4: "authored" ai generated content from prompts with or without human editing/review afterwards

Anyway... I'm curious if anyone cares or has options on the breakdown. This post would probably count as AI-2 "advisory". A new draft, written from scratch, but after running a previous draft past chatgpt for feedback.

 

[Veritable Quandry]

My students are allowed to use AI for research and topic development. Find sources, test counterclaims, improve their background. So a category to cover AI as a research and planning tool would be appropriate.

 

[Noodles]

TIL I write at AI Minus One.

 

[Jopsy Pendragon]

TIL I write at AI Minus One.

Ink/graphite on dead tree?

 

[Noodles]

Ink/graphite on dead tree?

No spell check and often no research.

 

[Argent Stonecutter]

I normally write in a plain text editor with no spellcheck or formatting.

 

[Jopsy Pendragon]

I'm a tragic case. Without vvo's built-in spell-check my posts would be a lot more embarrassing.

I can't remember the last time I used something that accepted text that didn't "helpfully" highlight non-dictionary words. Maybe the LSL script editor, but, well, that still has keyword/syntax markup.

 

[Argent Stonecutter]

I use nvi.