I feel like it would be good to train my own ai, but i have no idea what kind of business to make out of it. The only demographic I can think who would actively use it is you people...but then again...do i really want you people generating things...
Also, the computer necessary to run an actual llm is very expensive which is why i'd like to make some kind of money off it or at least try...but again, i have no earthly idea what kind of business people pay for with an ai...so idk, might just put it on here and see what happens.
Yah, I have to choose the level of training involved (tweaking, model changing, or from scratch)
The from scratch option is out because of cost (literally like 20K+) but I was looking at llamda and the specs are within my price range to get it going.
It would come with a preloaded model, so idk if it would be super racist, but you could probably bother it enough to call someone a nigger lol
That's the point. I've been working on a script on the shitter (I have an 8" laptop) to automate my invoicing from production (pdf), printing / emailing, posting in the ledger, etc. The processing power required is nothing. Total script may be 200 lines. PDF template(s) are about the same. CPU power required, one core pi zero could do it. Total time to create it is probably less than setting up and training an AI. Maybe four hours. Overhead: $0.
Define "learn." It can pull in data from my time sheets and ledger, calculate hours, fees, late charges, pull and populate addresses, etc. What's there to learn? Most businesses work from static data sets.
And don't get me wrong. AI can be cool in certain contexts. I did find some pretty neat art the other day (though most of AI art is shit). But it's all novelty. There's no real practical use, and again, because most businesses just need an accounting software or email that works, etc, but all of the npc options suck. Have you seen quickbook or outlook? Holy fuck. What a nightmare. Yet the same retards who brought npcs that shitware are going to save them with AI.
I could be the guy who said that no one will use a mouse, but I'm just not seeing it. Voice recognition / text to speech tech that isn't piped to google's servers would be cool, but you don't need AI for that. Sound in -> command could be scriptable too.
@Humpleupagus@Tony It will eventually have its use in mass data synthesis, but accuracy and reliability will have to make significant improvements first. Today, it's just Artificial Indians: cheaper call centers.
I'd love to for initially drafting letters, etc. Like a dictation service. The problem is that most text to speech isn't local as I understand. Android pipes the speech to remote servers, Samsung's or Google, and sends the resulting text back. Thus, it's all recorded by globohomo Corp.