💾 Archived View for tilde.club › ~winter › gemlog › 2024 › 10-19.gmi captured on 2024-12-17 at 11:09:33. Gemini links have been rewritten to link to archived content
-=-=-=-=-=-=-
Low-background steel (Wikipedia)
Project Analyzing Human Language Usage Shuts Down Because ‘Generative AI Has Polluted the Data’
Low-background steel is steel that has no trace of radioactive fallout. Since the detonation of atmospheric nuclear weapons in the 1940s and onward, there has been an amount of nuclear fallout that is traceable in everything, including steel. This is problematic because for especially sensitive cases (Geiger counters, spacecraft equipment), it's critical for there to be no trace, to avoid contaminating the instrument(s). Steel is made by forcing oxygen into pig iron, and so ay atmospheric contamination in the oxygen will also be present in the produced steel. And because of this, the sensitive steel has to be sourced by other means, such as by scavenging from shipwrecks.
I've been thinking about this in the context of Wordfreq, an open-source project that scraped the internet to measure the popularity of words in dozens of languages over time. A valuable tool to researchers, it is now, the team has announced, decided to shut down, the project no longer reliable because of the way generative AI spam has poisoned the online commons.
I'm sad, but I'm not surprised. AI slop is everywhere. Maybe you've noticed it too? Copy on a website that reads a little funny, like it's answering a prompt; searching for "raccoons" in Google image search, and how it returns a combination of real raccoons and lots of real-looking ones, covered in that AI sheen. The internet, our wonderful, awful, shared creation, has been taken over by a flood of generated text, images, and soon, video. You could never fully trust what you see - there's a reason "photoshop" exists as a verb - but to the extent that things were manipulated before, it was on a far lesser scale. There are no guard rails anymore, no constraints. Any idiot who wants to try to make a buck by flooding the commons, can; and will; and does. To say nothing of its other uses - states will use this to flood Facebook groups with disinfo, post fake videos to Twitter and Instagram and YouTube. The first age of the internet was naivety; the second age consolidation towards platforms, and this third, unending age, is using those platforms as a source for disinformation and hate. O that it were only spam.
The problem is one of responsibility, and the abandonment of such. But if we consider the ways in which our age is one of basic truthlessness, is surprise even possible? When there are no repercussions for lying, for cheating; when the dull and base are not ostracised, but celebrated; well, here we are, I guess. All the slop you can handle. Grab a spoon.
I first went online in the original sense of the word - on the (phone) line, connecting at 2400 bps to BBSs in my area code. I won't pretend that was perfect, because the very act of doing so kickstarted the worst year of my life. But at the end of every connection was a person, and it feels strange to miss that, even if with the good people, there were some very bad ones as well. But that's the situation we live in today: who's behind that avatar? That domain? Is it even anyone? Or is it an idler's plaything, a hacker's honeypot, a state intelligence service's scheme.
'I'm Making Thousands Using AI to Write Books'
People are Going to Die and Amazon Will Absolve Itself of Responsibility
AIs are coming for social networks
SocialAI: we tried the Twitter clone where no other humans are allowed
We don't know, we can't, and we can only find out by spending a disproportionate amount of energy digging to maybe have the sense of what something is. One good thing about the end of atmospheric nuclear testing is that background radiation has decreased to the point where the steel is now generally suitable for all uses. But this good news must be tempered (sorry) when taken as comparison: there is no 100% reliable test for AI text, and the genie is not going back in the bottle. With the rapid adoption of AI tools to produce SEO slop, and social network functions (hell, even your entire timeline on SocialAI), our corpus of text, whether electronic or print, is polluted forever. And there's no waiting this out. Having been given the tools, the lazy will continue to turn the crank. We call it crap; they call it creativity. But thanks to them, we all get to share in the consequences.