I'm in no way an AI hater.

[randulo@mastodon.social]

@nazokiyoubinbou Yes, Mastodon is a guaranteed anti-AI audience except for programmers who love using the tech but don't try to talk it up. AS I said, it's a technology and I neither hate or love it. The current implimentations like Gemini can do impressive things but are they worthy of attention? Not always. For example, it ingested a short story of mine and generated a two-person podcast discussing it. It sounded very much like two real people discussing the story. I didn't post that anywhere

[androcat@toot.cat]

@randulo @zebulonmysterioso

"The purpose of a thing is what it does"

[holzchopf@toot.community]

@randulo No, I mean in the sense that Mastodon is an AI sceptic bubble (tagirijus' words)

[bewilderbeast23@mastodon.social]

@randulo Pls name some good use cases. I need to know where I'm wrong when I think generative AI has none.

[randulo@mastodon.social]

@holzchopf Yes, no question about that! I don't think it's worth my time to hate on it publicly, even though I have the same objections most have. It is IMO neither as good or as bad as most people say. Two extreme fringes, and here it's the negative one.
Maybe more an echo chamber?

[holzchopf@toot.community]

@randulo yes it was definitely meant in the sense of echo chamber

[tagirijus@mas.to]

@holzchopf @randulo yes, lol.

[nazokiyoubinbou@urusai.social]

@randulo I think it's too lossy to be used for summaries. It's like trying to get the complexity of a concert with a recording on an old cassette from someone sitting in the back.

Most of the uses I can imagine up for them are more small time. For example, an author quickly testing out a concept to see how it feels in play. Everything I can think of would be fine with a local LLM and no need for server farms and compromising personal info for oneself and everyone one knows to achieve that. (People forget local exists.)

The true issue isn't even LLMs, but the way they're being used — forced into things, gathering data illegally, etc. Unfortunately, remove the LLMs and those behind them still do that stuff with something else...

And, to that end, the flex is absolutely negative.

[randulo@mastodon.social]

@bewilderbeast23 I did. Anything where a small artistic endeavor cannot afford human work but a machine could do it easily. Here's an example:
I used to use it for album covers. Music makes no money, and I can't pay a decent sum for an actual artist. Now I make my own covers, mostly my own photos. I already mentioned translation. Again, something I couldn't pay for.
One NOT good one is thinking it's your girl (or boy)friend. I hope that phenomenon disappears fast, it's dangerous.

[randulo@mastodon.social]

@nazokiyoubinbou Yes, local would be the way to go, but it still needs to be trained. If that can be done, it'd be great.

[heckinchonker@mastodon.social]

@hopeless @zebulonmysterioso @randulo In my experience they still need a human providing guard rails. In the hands of novice devs they often just enable massive amounts of slop.

[nazokiyoubinbou@urusai.social]

@randulo Bear in mind that a lot of what goes into the training is an obsession that they should be trained to act as a general assistant. ... -And that's the one thing they most should not be doing.

Long term, if people don't have a kneejerk uncontrollable hate towards the tech (which they might at this point thanks to all the abuse) I would say the best solution is a community run actually open model where stuff that is actually free and open and submissions that can be appropriately licensed, etc etc are used for training. It would be a slow process that would never produce something as fake-smart as the current models — and that's a good thing in its own way because it should not be used as a "general assistant."

As it is though, when the bubble pops, the tech likely will die.