the on-going AI class divide
Currently thinking about how using AI will just increasingly become even more of a class/privilege thing than it already is. We've already established that different groups and areas around the world get access to models differently, but one puzzle piece in the AI discourse is obviously money.
If I have the money (and even the willingness to financially support) OpenAI with a subscription, I get a much better experience. Aside from inflated claims for marketing and employability, there might even be a little bit of truth in it when an AI enthusiast says it did a job well and saved him hours of time - but likely just because of the subscription model. More tokens, more uploads, deeper 'thinking', better search, earlier access to better models and an interconnection between software matters.
As an example, one AI use case I could appreciate: Summarizing my existing study notes for this semester, and giving a brief summary of each court case I mention as source or reference, just as a reminder of what it was about. But the uploads, 'thinking', web search and output size are strongly limited on a free plan, and it would have to read at least 20+ lengthy court cases per note, and there are at least 30 .md Obsidian notes to upload. Half of what I want isn't possible, and uploading all my notes when I am allowed 3 per 24 hours will take me weeks.
I imagine my use case would be less of a problem in the upper subscription tiers, even if spread out across a week, but I am not willing to pay that. People using the free tiers spend a lot of time waiting out cooldowns, repeating and rewording lengthy prompts and searching for ways to make happen what the subscription would enable instantly.
That makes discussions with others around what AI can or can't do unreliable and mostly uninteresting to me at this point. There is too much variation of output between the same prompts, too much difference between different model releases, and the intense advantage of paid tiers. Nothing novel about discussing why a free product is worse than a paid plan.
This also isn't me complaining about why something with higher quality and more resource draw is more expensive, that's a no-brainer. I'm just focused on the fact that without being upfront about the setup and costs, comparing notes is futile. I don't want to hear about your great setup if you don't also mention that you're employing 4 LLMs at the price of 250 dollars a month, that are all integrated into your workflow that features 4 other software subscriptions. On top of that, I see this working for a lot of tech, but it sucks in academia, because academic access costs thousands and you rely on your university's subscription and proxy that AI can't use for searching.
I fear many of you are no longer in touch of how the logged out, chat-only experience is. Reminds me of my workplace that sends us to AI classes where the Enterprise license is used, just to give less than 10% of employees actual access to the Enterprise version at work, meaning no one can use what they learned.
I anticipate this divide to get worse as the companies are nowhere near profitable and will enshittify at lightspeed, especially on the free tier. A significant amount of ads, frequent cooldowns, and even harder upload and character limits than now. Just good enough to get you hooked a little, just shitty enough to at least get you on the 8$ tier. People who tend to be left behind and too busy working long hours or multiple jobs to stay afloat are being goaded into thinking this money is the only thing left between them and a better life or more intelligence.
So it's worthy to consider that if someone has an absolutely garbage experience with AI that you can't understand, it might simply be because they're poor.
This of course leaves out the existence of FOSS LLMs you can run yourself for free, but not only is the setup a bit of a hurdle for many (in a world where more and more people do not own a PC), but I don't expect the same resources, fine-tuning and websearch from them, so in my eyes, it doesn't really make up for the disadvantage.
Reply via email
Published