If it was literally impossible to make any profit off AI, how do you think things would look like now?
AI existed long before Big AI. Even if there was no profit incentive, it would still be useful for academic research, physics research, language studies, etc.
LLMs were invented in 2017 (as an improvement upon RNNs from 2014) but for-profit Big AI companies didn’t really exist until 2022.
Note: OpenAI came out with GPT-3 (175 billion parameters) in 2020 but at the time it was still a non-profit company. That year was the point where Big Money decided to invest heavily into AI.
zero profit incentive [for a technology that costs $billions]
It would not exist. Or if you insist, it would be a science-fiction book, Asimov wrote a lot of that.
It wouldn’t exist outside of a niche hobby for enthusiasts.
Niche hobby enthusiasts or altruistic people
Altruistic or autistic?
Yes
It’s not really a hypothetical when no one is currently making a profit off of LLMs.
Plenty of slop is making a profit passing off as news sites. There’s also the video slop flooding the big social media sites, earning ad revenue
The more of them that are, the less they’re worth. It’s only a matter of time before it costs more to make a slop video than what it can make in ad revenue.
I don’t know the pricing, but the cost is going up while the revenue is going down. The easy slop earning model is going to peak within a year if it hasn’t already.
Maybe the open source stuff would exist, after all there will always be nerds doing this stuff for the fun of the game, and some companies would be using the open source models to build customer chat bots or in-house troubleshooting agents. Things that help them move faster or more efficiently. I don’t think the audio, video, or code generating models would be far enough along to be useful at all.
Dont the companies make huge losses constantly with the llm crap, and the whole thing has been just speculative with hopes of everyone hopping in on the llm train and being dependent on the corporations that provide them?
The LLM providers are making losses, but that doesnt mean no one is profiting. Cyber crime, social media revenue, propaganda networks. Lots of downstream users of LLMs are making money.
Well yea, but its very fleeting if the foundation is crumbling. And good thing it is, if the providers were getting what they imagined they would we would be fucked beyond belief.
The tech won’t go away though, the models will just be hosted on local GPUs. The audio and video generation might get more expensive or out of reach for most of us, but the image and text generation will be available to anyone with a basic GPU.
And sadly, cybercrime, propaganda etc are all backed by very deep pockets, so they may be able to afford to pay the real prices.
I dont think the genie is going back in the bottle sadly.
At least it isnt the worst case scenario, where most are addicted / chained to llm usage, nothing can function without it and where corporations could influence what users think on a whim by instructing the llm to act in certain way. Imagine if everyone was as crazy as some of the ai psychosis persons are and those who are not were considered to be the deranged ones. I hope we managed to avert that nightmare.
afaik when GPT-3 first came out there was still no profit incentive, so maybe something like the new mistral model would be the frontier today. Just guessing though
I have no clue what you’re trying to ask
Since the majority of AI use today is fully profit driven, what would be left if there was no money to be made at all with it
A lot of people/companies that use it do it not to directly make money, but for productivity gains and automation that saves time or money or finding issues they didn’t know they had before other people do.
It’s like saying “what if there was no profit from using this chainsaw over a two person bow saw to cut down a tree.” I am not directly profiting off of the chainsaw, but it saves me paying a person and allows me to be more efficient.
The only people “making money” from AI are the companies hosting it, and there are plenty of local models you can run instead, they just might not be as good/scaleable yet.
Less focus on NLP and Computer Vision, and more research in other subfields of AI.
