Discussion
One of the main objections I have to "large model" AI is that it is a form of piracy that fails to serve as an accurate form of archival. https://vis.social/@bleeptrack/117560078386333325
Every AI scraper develops an archive of copyrighted works that would put any library on earth to shame. Then they only let you access it it horrifically garbled, with semantic noise added, and through a straw
@allanb in the long run i think this will be bad for preservation. with a romhack, you changed the original, but you know what you did. "AI" recompile artifacts constitute randomly generated false versions of vintage games where the developer does not and does not have the ability to actually know what exactly they changed, and to the extent they do know is potentially not honest about what or how much they changed.
@mcc This particular game has a huge amount of 3rd party community maps that were painstakingly created by hundred(s)? of people over the years. The maps can all be re-used in the AI version, but are almost unrecognizable due to "aesthetic" changes added-in by AI. IMO optimizing a vintage game to modern computers is one thing -- changing it's appearance and play is another thing entirely, defacing original art. I doubt the AI devs can separate the two products.
@allanb The important thing is that you are featured on Reddit.
@mcc I really liked Ted Chiang's framing of this in this article: LLMs as a form of lossy compression, a "blurry jpeg of the web"
https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web
@mcc it's just wild to even think about the scale of data hoarded by these greedy companies only to gatekeep it behind a restrictive API or chat interface that actively takes away the context of original works by talented people.
@mcc Nailed it. We need laws against this.
The (imo) clear sign that this is intentional is that LLM interfaces are offered by those powers that disable access to the non-garbled indexes.
This begs the question of the goal of making access to public information harder.
The only answers I can see are ethically appalling.
Regardless of the question of the use of llms, the greater good of disabling free direct search is obviously non-existent.
@mcc first they steal it, then they go bankrupt, then they sell the data for pennies
@mcc Also, when they scrape the real world, i.e. books, magazines etc. they FUCKING DESTROY THEM IN THE PROCESS, which I'm pretty sure is the opposite of preservation...
@mcc That's why you feed whatever LLM you use garbage on purpose so their data becomes meaningless and by doing so no one will want to use it. #PoisonTheData #PoisonData
@Wiz4rd_of_Aus Theoretically anything you feed into the LLM goes into the context window, not the training set. Unless the AI companies are lying to us. Would AI companies do that? Just lie?
@mcc That's what I mean. Do it anyway incase they are lying.
@mcc what if a bt client that always trains a (toy) transformer model as it downloads anything, this ought to now have a lot of legal status now right?
@PsySal
I know adding a toy transformer model to a BitTorrent client is not a serious suggestion, but let me pretend for a second it was! 
There’s two major differences that make this non-comparable, really:
Sooooo, not really the same no.
@mcc
@curiousicae @mcc fair commentary although I was pretty sure I read that they did not just leech but were also uploading...
@PsySal @curiousicae The actual flaw in the joke plan is that running the bittorrent client may turn out to be the ONE THING that the AI scraper companies actually get in real trouble for.
But not enough trouble. It will be a fine.
This means that the real hack to being able to commit any crime you want is not LLM training, but money
@mcc look at us, preserving these books. we're preserving them so hard that we're physically shredding them!
what? no! we were FORCED to shred them, by this ASSHOLE judge who said it would be copyright infringement otherwise! blame HIM, not US, who could have simply not trained on these books and chose to shred them anyway!
@mcc this clicked some things into place in my brain hard enough that I want to scream now
@mcc@mastodon.social It's frustrating to find stuff with too given how it works internally and dnesn't understand the material. I try and search up a schematic and the google AI I can't get rid of wants to direct me to useless places with made-up answers
@Selinica I've switched to another search engine where the AI has an off switch
Chokepoint Capitalism mind melded to Surveillance Capitalism and fossil fuel funded technofascism.
@mcc Must continue to pretend that it's not theft!
@mcc LLM, is I stole your work, and now you need to pay it, to access it!
![[empty]](https://files.mastodon.social/accounts/avatars/106/887/462/825/018/571/original/52e5d4364f73668f.jpg "[empty]")
@mcc It's a vexing problem with modifications of "dead" computer games too. People wrap the game with AI optimizations, then add in modifications over the original work, and basically combine both into a new game based on prior work ostensibly without permission of all who created it.
(example: https://openomsi.org/)
openOMSI