GenAI bad

First things first, I’m not a huge fan of copyright, especially as it relates to corporations, and especially software they create (which also has additional restrictions to make sure you don’t break thier precious totally breakable locks). I think it’s culturally restrictive, it’s made for a capitalist system, and it obviously favors large entities which hold it.

The recent flood caused by genAI has caused a resurgence of pro-copyright takes, and I feel corporations that benefit from the copyright regime are definitely having a moment right now.

Copyright is a blunt tool

The laws around fair (or allowed) use vary from place to place, and are usually pretty specific with the kind of use allowed. They cannot currently distinguish between uses of machine learning that would be considered good or bad, fair or unfair.

One could make the argument that training data should always be collected with explicit consent, however this is impossible to scale, and we have legitimate uses for ML that in my opinion shouldn’t be thrown under the bus. Translation, voice and image recognition, generating text descriptions, and all sorts of other things that don’t directly compete with the work being used for training and/or make the world more accessible. Not being able to use public data would also particularly screw over small and open source projects.

GenAI is evil no matter what

In my opinion it really doesn’t matter how the data used for training genAI is sourced. You could make a deal with aliens to synthesize the images or words for your dataset using the power of their immense intellect, the end result is the same - planetary scale spam. This is - I believe - the main mechanism of harm when it comes to genAI, we are not capable of expending the effort to weed out all the sludge that comes out of these machines and floods our digital spaces, and this in turn hurts those who rely on them, which to varying extent is almost all of us.

The direct displacement of people due to genAI is sadly also happening, however there is one small problem which means it probably won’t last long-term…

It works! …until it doesn’t! This technology is severely overhyped and unreliable for any legitimate real world usecase that requires any degree of certainty. Despite this, it’s being crammed into every possible place, and inevitably, fails.

This really only leaves the use of genAI to bad actors and as a curiosity.

Let’s ban it!

Yes, this is where I call for a blanket ban of a technology!

We need to recognize the harm genAI is causing to the internet and people who depend on it, how it’s being used to spread disinformation and making people unable to trust anything, how it gave a bad name to the field of machine learning as a whole, and finally how it made the already shitty job market even worse due to unhinged business people wanting to get in on the hype. I think the only option is to put a ceiling on how much compute anyone can use to train these models, no exceptions for the big guys.

The not so great part about it is we can never be sure that powerful instances of this tech do not exist within government agencies, or even secret GPU farms. This is true no matter which way we go about fighting genAI, the bad actors won’t play by the rules. We also have models released to the public that we will never be able to erase.

This is scary. There’s a lot of power in being able to fabricate things that do not exist, at scale, and it puts the dead internet theory into the plausible realm. It makes us retreat into the dark corners, hoping to maintain genuine communities, as long as it’s still possible.