General AI thread

[gmoney]

6 minutes ago, Sphinx said:

I'm sure there's some interesting points in there but their writing style turned me off. Can't help thinking their writing would have been improved if it was put through ChatGPT...

You are sort of established as a bit of an idiot around here though, so I'm not sure your opinion counts for too much. 

EDIT: that is a bit harsh, but it is important context. 

Edited June 25, 2025 by gmoney

[SuperBacon]

Someone declined a meeting with me 5 minutes before it was due (super annoying) but then tried to send in some sort of AI meeting bot to take notes or some shit (I didn't let it in the meeting)

But the meeting was just me and her, so what was the bot taking notes of exactly?

[tiger_rick]

17 hours ago, SuperBacon said:

But the meeting was just me and her, so what was the bot taking notes of exactly?

Bacon is having a shit.

Bacon is eating crisps.

Bacon is reading aloud from a British National Front website call the UK FF.

[JNLister]

Long story short:

• LLMs are effectively the autocomplete from your phone's text messaging, but a zillion times faster and instead of using your message history to predict the next word, they use a massive database that can literally be "the entire internet and even more".
• LLM's don't "write" or think. They use the prediction to simulate the writing process, specifically the outcome. They don't try to write a letter of complaint, they try to produce a document that looks like what somebody who was writing a letter of complaint would produce. 
• Weirdly it turns out this simulation process can have some practical benefits and uses. In particular, it can give the effect of better understanding and acting on a complex query or request than search engines in some cases. You have to understand what an LLM is doing (and is not able to do) to make the most of it. Actually having it write a chapter plan for your book is dumb. Using it to see a simulated outcome of what a chapter plan for your book might look like, because that's useful input for you and gives you ideas (or throws up something obvious that you'd forgotten), is not dumb.
• Many "uses" of an LLM are ineffective at best and harmful at worst. If you think it's actually writing/reasoning/researching, you're probably going to get the worst outcomes. If you think you can ask it to write something and then treat the output as if a skilled human had written it for you, you've missed the point. 
• Completely counterintuitively, the best way to get good results is often to "talk" to it as if it was a person who was actually doing the task you wanted, rather than a software tool you are trying to correctly program. It's a weird kayfabe that only works if you know it's faking what it does but pretend you don't.

tl;dr: An LLM doesn't write stuff. It simulates the outcome of the writing process. If you understand that, you might find that simulation useful. 


 

[The Gaffer]

3 minutes ago, JNLister said:

Using it to see a simulated outcome of what a chapter plan for your book might look like, because that's useful input for you and gives you ideas (or throws up something obvious that you'd forgotten), is not dumb.

It may not be dumb, but I'd purposely avoid a fiction book whose narrative I knew was stress tested and game theoried in an AI environment so that certain beats could be hit rather than the a̶r̶t̶i̶s̶t̶ content creator using their conscious intuition and imagination to get there themselves. 

I appreciate that something similar is probably done with one-a-year crime thrillers etc. I stopped reading Brandon Sanderson's fantasy novels too as they seemed to just strike me as pure, relentless content generation spreadsheeted to give the reader what they wanted. 

It's a personal thing - that kind of entertainment is fine - but not for me. 

[LaGoosh]

29 minutes ago, JNLister said:

Using it to see a simulated outcome of what a chapter plan for your book might look like, because that's useful input for you and gives you ideas (or throws up something obvious that you'd forgotten), is not dumb.

A good alternative to this is running it past another human being whose opinions/ideas you trust and respect.

[Sphinx]

I gave the rest of the article a read and I enjoyed it. I was maybe in a silly mood, genuinely had the thought cross my mind that the writing I'd read to that point could be improved by ChatGPT, and found it funny.

I'd written a longer post to this but, long story short, although I've known there are downfalls to AI, I've been reluctant to completely shit on it as I've also seen positives in my life with it, and have been interested in how it could be used as a tool. The article does a good job of outlining the dangers even more clearly and I'll check out some of the referenced links at some point.

Though I'm interested in how it can be used as a tool, overall the cons outweigh my curiosity.

Edited June 27, 2025 by Sphinx

[d-d-d-dAz]

https://www.stereogum.com/2313501/ai-generated-psych-rock-band-the-velvet-sundown-rack-up-hundreds-of-thousands-of-spotify-streams/news/

Apologies if this isn't the general 'Fuck AI' thread, but this is horrifying.

It looks as if the main publishers might have worked out that AI bands mean saving money on royalties, and now we have an entirely AI generated band - with accompaying promo shots and social pages - being actively pushed via playlists on Spotify. They have 00s of thousands of streams.

Depressing.

[LaGoosh]

It was inevitable and we're going to see more and more of this. This is what happens when the arts get taken over by tech bros and private equity. The next step will be concerts where people watch AI bands procedurally generated virtual avatars on screens playing AI music. Crowds will be able to interact with the "band" live via an app on their phone.

[johnnyboy]

2 hours ago, LaGoosh said:

The next step will be concerts where people watch AI bands procedurally generated virtual avatars on screens playing AI music.

Abba Voyage was a warning shot.

[BomberPat]

On 6/25/2025 at 6:19 PM, SuperBacon said:

Someone declined a meeting with me 5 minutes before it was due (super annoying) but then tried to send in some sort of AI meeting bot to take notes or some shit (I didn't let it in the meeting)

But the meeting was just me and her, so what was the bot taking notes of exactly?

My team are moving towards AI note-taking; part of our remit is minuting and supporting committee and department meetings, which doesn't really fit with the rest of what we do and we just don't have the capacity to do it to the extent that it would be worthwhile. So our boss has been pushing for us to not attend the meetings, and just use AI to minute them. Problem is, we only have two licenses for the AI software, so one of the two people who have one have to be invited to every meeting. And then the notes are really no better than just letting Teams do a transcript anyway; it tries to organise things into vague topics or talking points, but unless you really mechanically state "this is in relation to agenda point 1" or whatever, it just makes it up. And most of our meetings are hybrid, and it's useless when it comes to that.

[SuperBacon]

One of my colleagues passed her apprenticeship this week, and another colleague (and very good friend of theirs) wrote an email to the team congratulating her, but obviously used CoPilot to write it. 

Can people not take 5 minutes to write a heartfelt message for something that requires it? It's so soulless. 

I bet people use it to write wedding vows and the like.

[HarmonicGenerator]

1 hour ago, SuperBacon said:

One of my colleagues passed her apprenticeship this week, and another colleague (and very good friend of theirs) wrote an email to the team congratulating her, but obviously used CoPilot to write it. 

Can people not take 5 minutes to write a heartfelt message for something that requires it? It's so soulless. 

I bet people use it to write wedding vows and the like.

I found out this morning a friend from uni had passed away. His sister-in-law had posted about it on Facebook, and when you clicked on the post (to leave a message or comment) it pushed Meta AI on you to write your note of condolence. I know in these situations a lot of people can't think of the right thing to say, but it just felt so impersonal and not right.

Edited August 8, 2025 by HarmonicGenerator
I'm aware posting a message of that kind somewhere like Facebook is in itself quite impersonal

[DavidB6937]

Someone mentioned the same thing to me when they were writing something about my son passing. Absolutely ridiculous. I'd rather people said nothing than some AI shit.

[JNLister]

On 8/8/2025 at 6:31 PM, HarmonicGenerator said:

I found out this morning a friend from uni had passed away. His sister-in-law had posted about it on Facebook, and when you clicked on the post (to leave a message or comment) it pushed Meta AI on you to write your note of condolence. I know in these situations a lot of people can't think of the right thing to say, but it just felt so impersonal and not right.

My mum hit the Meta AI button in WhatsApp without knowing what she'd done (or what it meant). She then exchanged five messages back and forth thinking she was still messaging my sister before she realised what's up. She is now not a fan of AI.