I was going to say I don’t, but actually I do use it in my classes for one specific purpose, that is, demonstrating to students why they shouldn’t use it. Generally, I do this by asking it to tell a joke about three random items (picked by students) and then to explain why the joke is funny. At least one of two things invariably happens: it comes up with a joke-shaped object that isn’t actually funny, or it sticks close enough to an existing joke to come up with a reasonable punch line, but then muffs the explanation.
I just did that with a ruler being the object and it did fine, IMHO:
Chat GPT-4o
Joke telling, though, is one of the weaker aspects of Chat GPT. Normally I have to ask it for ten variations to find a few that are workable. This time it got a reasonable joke on the first shor.
I remember asking Chat GPT 3.5 to come up with a Flintstone-style punny version of “Donald Trump” and it failed miserably (just giving some lame rock-related puns that had nothing to do with a name).
Yeah, there’s a gulf of difference between 3.5 and 4o. 3.5 was definitely cool when it came out, but, in creative tasks it’s gotten so much better. Like a human? Well, I’d instinctively say heck no, but, honestly, it outperforms a good portion of the population. I’m literally using it right now to help work out some final stanzas of a song. I’m using none of the actual lines it suggests, but it helps to organize what I’m going at thematically (I often don’t know exactly what the hell I’m writing about except in a vague way – words and phrases come to me that sound good, but have meaning that I can’t quite put my finger on.) It helps in this way, and it helps in suggesting what directions I could go to complete a thought. When I’ve asked it to analyze the current song, it does a damned good job, almost like a good therapist, in putting together what I’m getting at, and helping me understand what in the fuck is going on in my brain as I’m writing it.
I still wouldn’t trust it for anything more than a middle-school or possibly slightly higher level poem or song, and it’s certainly not better than what a bright student in that range could come up with. But it has improved leaps and bounds over what we had in November (or was it December) 2022 (which I still found unbelievably impressive.) I’ve always seen literary creativity as among its weakest points.
I’m curious, as a songwriter, have you tried (and do you have any thoughts on) suno.com? They’re basically the ChatGPT of music – one of the very few actually-differentiated AI models/services out there, in fact, as opposed to the millions of generic chatbots.
It is specifically focused on music, and generates entire songs (melody, background, lyrics, album art, etc.) given a prompt. You can give it a subject, style, genre, etc.
As a layperson (meaning non-musician) I find this super impressive, but I’m not at all qualified to determine whether there’s anything, er, legitimate “musical” about it. How would it compare to your average “bright middle schooler”?
Here’s a fun (if somewhat nondescript) one it made about the SDMB 
https://suno.com/song/9f06fd9c-5e45-4384-b232-2dad9e59a1b2
wow, color me impressed …
I’ve mostly used Udio, but started out with a couple spins on Suno. Anyhow, all the same idea. (I would consider myself much more an instrumentalist than a songwriter, but I dabble in various kinds of composition.)
My opinion? The first time I tried it, my jaw almost hit the floor. I had no idea we already had this kind of technology. I would have been happy to hear a coherent MIDI file of a prompted composition. Here we have full orchestration, including vocals. Holy crap!
I personally think it does a good job. I really can’t objectively say otherwise. It’s sometimes a little predictable (but what pop music isn’t), but it’s thrown curveballs at me. It’s musically coherent, well arranged, well orchestrated, etc. I haven’t used it for any riff ideas or anything yet, but it’s certainly given me ideas about stuff like the vibe of the song and the orchestration. As for “bright middle schooler,” I think it’s significantly past that level.
Hadn’t heard of that, thanks!
I got some strange results out of it… using the same prompt as with Suno, I got some weird incoherent gibberish that sounds vaguely like human speech, and doesn’t match the supposed lyrics (until the chorus): https://www.udio.com/songs/gAo4bjuueE526RtYEjbyLF
Wonder if I’m using it wrong, or if maybe the paid version is better?
Suno’s generation seems much clearer, otherwise (and the songs are longer and seem to flow better, but that’s strictly IMO).
Well, damn. I wonder how many kids it’s going to turn off music 
Or hopefully they’ll just learn to live with it like we have to…
Whoa. That is really odd. I’ve never had it do that. I have had it occasionally miss a word or change it for some reason, but that like, I dunno, one in twenty times?
I’ll have to see what it is like now. When I was using it was about nine months ago, and Udio was significantly better than Suno to me. Maybe Suno has caught up or surpassed Udio. I’ll have to try it.
I mean, I don’t know. I’d like to think not many. The desire of an artist is to create for creation’s sake, at least that’s how I’ve thought of it myself. I mean, many people want recognition, but the reason I make music or write things or cook or draw or knit is to create something that wasn’t there before, that was the result of me (or a group of people). Let’s pretend AI can 100% create music that a human couldn’t distinguish from human creation and is as interesting as what an exceptional human can do. That’s not going to stop me from creating. Those aren’t my babies. And I hope there’s enough kids and people in general who feel the same. I mean, why else pick up music?
Great point.
I feel the same way about writing (English)… ChatGPT can already do it better than 90% of people I know, and better than me half the time, but I still like doing it on my own It may be crap, but it’s MY crap, dammit!
Apparently it’s a known issue with their model. I made a bug report and this is what they said:
You may see Udio adding vocals to your tracks sometimes, these vocals mostly have no proper pronunciation of actual real words. These are often referred to ‘nonsense vocals’, ‘hallucinations’ or ‘Simlish’ and are an inherent part of the current Udio model. It’s possible in future updated models that their frequency may be decreased.
I guess I can live with my Sims singing in their native tongue from time to time. Good practice for Skynet: The Musical.
I meant to get back to this thread earlier. The last creations I had on Udio were from eight months ago. After you mentioned your issue, I tried creating four tracks, and one of them had the goofy error your had, and one other just came out as complete jibberish – the “Simish” you mentioned. That’s weird, as I have about a hundred songs in my library, and not a single one has “Simish,” though a few have words inserted or changed. It seems like the current model is buggier than the old one. Huh. Will have to check out Suno soon.
ETA: Just checked out Suno and, yeah, it seems to be much better than Udio now.