@sidereal That explains a lot honestly. I was never big on video and couldn't really understand why so much social media was centering it as a service and platform but I thought I was just an outlier, this fundamentally makes more sense to me

@sidereal
> What makes AI any different than any of this shit

AI serves the narrative that it can be used to get rid of humanity; A primary ambition of the ruling class.

Money matters a lot less than the possibility of killing off billions of people. Especially if a computer system can do it

@sidereal @ripperdoll
A lot of people lost their jobs over it. Cracked had a lot of really good talent that they had to let go when the pivot-to-video stuff all collapsed.

@sidereal that's quite a list!

@sidereal

I guess L.L.M.s are uniquely great at helping the surveillance capitalism industry. Google was already scanning every body's emails, but now they can do deeper sentiment analysis and automated summaries and build more detailed dossiers on every body, I don't think they will ever give up such a shiny toy, and Google doesn't give a shit if it makes mistakes and generates inaccurate biographies about us

@sidereal

As far as I can see among my circle of friends, there _is_ a demand, such as generating aids for a #ttrpg session.

But I doubt that this demand would be sufficient to generate the trillions in revenue that the AI industry claims will happen any time soon.

@juergen_hubert @sidereal Given that the whole TTRPG industry is smaller than Candycrush, I doubt there'd be money even for a smaller image-only startup there. People will just go back to Google Image Search and random tables.

Image generation outside of this sector might even be a profitable non-insanely-huge-datacenter XXXX-LLM use case, as when the current status quo breaks down, it'll be due to financial, not legal issues, sadly.

@mhd @sidereal

Isn't the expensive part training the LLM systems, rather than running them?

@juergen_hubert @sidereal Well, one is a lot more expensive, but that doesn't make the rest really cheap. So offering servers for that still makes sense -> easy enough to sell a subscription.

The "good" thing about images is that they get out of date less fast. With text, it's bad if your knowledge is cut off in 2024, but when you need some image for an article or presentation that shows the pyramids with some modern building equipment, you don't care too much about that.

There already are a lot of companies doing that and not all just piggy-back onto the big models as is the case with 95% of AI startups, so I would think that's a sector that could survive the big players' bubble bursting loudly.

Someone doing that just for fantasy/sci-fi/TTRPG images seems less likely, as there's just not enough money in that.

If DMs already got really invested in "creating" all kinds of crappy scenery image for their games, they either have to switch to modern scenarios or maybe furry games (because, wow, there's definitely money in that art sector).

@sidereal i saw a screenshot of a company guy that was freaking out because they have been using AI to summarize analytics data and make executive decisions about consumer preferences

And they just realized it was making up everything this whole time

@mhd @juergen_hubert @sidereal You don't even need an LLM for image generation, just a latent diffusion model, and there are small, fast, efficient open source ones everybody can download and run on their own GPU. Generating an image doesn't need any more energy than running a high-end PC game with the graphics dialled up to eleven for a minute or two.

@LordCaramac @mhd @juergen_hubert I can generate an image of a DND character even easier than that by drawing them on a piece of paper!

@sidereal @LordCaramac @mhd

Not everyone has developed this skill.

I mean, _I_ wouldn't use AI tools for that, but I can see how it would be attractive to a lot of people.