Recommended listening: Virtual Insanity - Jamiroquai
You do not have to start out every anti AI statement with "it has uses" or something else along those lines.
You can just outright say what is happening without having to twist it into being "well actually both sides have a point" bullshit.
Yes, on a fundamental level, the technology that is LLMs could be used in beneficial ways, as demonstrated through Alphafold being implemented to predict
structures of proteins of covid. This method had previously recorded protein families as training data, and was made open access, so anyone can have custom versions
of it for non-commercial use.
Did you pick up on the differences between that implementation and the heavily pushed models being sold by snake-oil salesmen?
First of all, there's the issue of training data. To ensure a model is the best it can be for a specific task, it's imperative that the training data it's built on
is entirely related to what it's being used to support. Any non-related input can lead to inaccuracies in the desired output. So, what do you think the use is
for models that guzzle up everything publically available on the internet? Or how about models that actively destroy real books because it'd take too long to not destroy
them? They're trying so hard to be all-purpose software, but in doing so they're actively making their product worse. They have to do this however, because how else would
software as dedicated as it wants to be attain a valuation as large as it is? If it isn't able to do your taxes, graphic design, coding, be your lover or anything else all
rolled into one product, why would people not just go to the places that can do it better (that are also dedicated to each individual task)? The only percievable product
for a model trying to devour the entire written word is something that just seeks to replicate it, and the main application of which seems to be just something to talk to
(that'll probably try to murder you, like models on the market now are currently doing). Genuinely insane that so much money is being put into the murder machine, is the
graduate job market not already doing the same thing for less?
Additionally, another thing that Alphafold did that these pushed models don't is that it's designed to run locally. This method only really works for specialised cases,
like our example, so it wouldn't work for a model with way too much training data to bloat it. And the thing with that is it's a lot more difficult to monetize software
that works this way, not relying on a connection to the internet to function. Given how shitty the big AI companies' security seems to be (looking particularly at the
latest anthropic case of privately shared chats just being publically available via search engine), probably not impossible to be able to pirate their software, but a
lot more challenging than if it was just an executable on a local system. They need to sacrifice utility to be able to make a profit (even though looking at their
financials making any kind of profit is an impossible task), but in doing so they fundamentally misunderstand the tech.
As an extra note because it's another thing I see around a lot, if anyone tries to sell you on "agi" being a concept achievable within the decade, you should call
them out on it, because it's a physical impossibility with large language models. Artificial General Intelligence is a delusion, built on the desperate hope that all the
money thrown into the chasm that is ai will someday be worthwhile, but the truth is that such mindless frivolity is a disgrace to everything and everyone.
If current products are being made to fix actual problems, this is not the way they would be implemented. They feed off the good publicity that helpful projects (that
aren't related to their business model) produce, and use it to further drive data centers and tracking technology that exist to worsen quality of life for everyone except
those wealthy enough to be above the system. Do not allow them to feed on your words too.