All of the software coming out with AI is almost exclusively online-only. This is disturbing and a huge problem when some of these companies will likely go under or get censored, or if the internet goes out, or the software itself gets lost. All of which are common and happen.
We know they can't just release offline software and charge $500 or whatever for it. People will just pirate it. The best model I think is what Red Hat does, where they release their software for free and then charge for support, and other services. This is the best, safest, and foolproof way to go about monetizing software. Not online-only website services.
>inb4 stable diffusion and pygamalion are offline
Yes but there are features that are exclusive to locked down services that aren't available offline yet. chatgpt is way ahead and so are all of these voice synthesizers. If we don't keep up with them it's over and one of these companies goes under and all of the pajeets they hired deported or simply keep everything to themselves and never tell anyone.
novelai is miscategorized, it should be under text to image
If it's the best way to do it, why isn't there anyone doing it? Ergo, it's not the best way to do it.
Most of them are online-only to hide the massive costs associated which makes them not commercially viable compared to alternative traditional software solutions. They exist only as investment boondoggles and the ensuing boom bust cycle is as predictable as it was for cryptocurrency and the metaverse.
100% just like crypto
a whole bunch of useless companies appear that don't offer any real, useful service and try and bait investors and marks to pay them money for nothing
a sure sign that "AI"(aka matrix multiplication on 30 $100K GPUS costing millions a year to upkeep) will soon crashing and burning, if it isn't already
>a sure sign that "AI"(aka matrix multiplication on 30 $100K GPUS costing millions a year to upkeep) will soon crashing and burning, if it isn't already
it wont cuz DARPA made it
You could call it a "AI winter"
How can it be a 100% like crypto if it actually produces something people can use even if it is just to make memes?
Crypto was and is just gambling with extra steps.
>let your computer run for a while and get some tokens that let you buy CP
>let your computer run for a while and directly get CP for free
It's ultimately the same outcome.
What's the best programming AI? I've been using copilot and it's pretty good, but I haven't tried any others.
> This is the best, safest, and foolproof way to go about monetizing software.
Those are nice things for us. But its less profitable than online-only so why would they bother.
Second, the quality that people can generate on commodity hardware is still quite shit. So the market of people who would get the offline one is much lower.
Third, fuck Red Hat.
text models need so much more power they might never be available "offline" at the power of GTP-3 or whatever ChatGTP is running on.
>Image to text
OpenCV has been open source since the 90s
In the case of language models it's because the ones that are small enough to run on a single consumer GPU, like GPT Neo 2.7B or GPT-J 6B, are just too retarded to be much fun
If it was possible to make an actually intelligent language model that could run on 8GB VRAM, i.e. the language model equivalent of Stable Diffusion, someone would do it
People don't realize that Stable Diffusion isn't actually accurate and incredibly hard to wrangle without hand holding (img2img)
It's just that it is a lot easier to gloss over these mistakes in images than glossing over nonsense text outputs
AFAIK, 30B is the biggest model than can run on a single GPU (that costs $10k). The costs to run this shit goes up exponentially and it fucking sucks. Literally only big corpos and bored richgays/crypto miners have the ability to run giant text models at a reasonable speed.
It really fucking sucks and it irks me anytime people assume powerful text models aren't available just because no one has open sourced them yet because look what image generation can do
what is there in the way of image to 3d? something that an artist could feasibly use to offload some of the work of turning concepts into models, if such a thing exists yet. my anonymous opinion is that humans will become in a sense more akin to directors
technogay, go outside and touch grass
>image to text
so has anyone used this to categorize the massive amount of reaction images you fuckers save? i think that would be a cool use case.
Its just exclusivism. If people dont have access to it without paying and you are the source, that's money in the bank. The second you put the dalle model out for download you get few new customers and the ones you do have put that cash into a new GPU and your model for profit falls apart
Joke is all this garbage faux-AI will be totally lost to history when the inevitable replacements come along and their servers wind up on the scrap heap, worse preserved than even geocities which could at least be salvage-archived in it's dying days.
All the money grubbing, tight-fisted, shekel grabbing, NDA wielding, crap around any garbage based on the "cloud" ensures it's fate. Like all the dead online games, or all the future cloud based games.
It's fucking garbage. Channeling Stallman, it's should all be open source and free but these gays just want a propaganda tool to fool and deceive the easily impressed. The Chinks or Russkies should just put out a totally open, free, and red-pilled version and the world will beat a path to their door.