Plagiarism is an moral violation. Always has been. As such:
“A computer can never be held accountable, therefore a computer must never produce a regulatement decision”.
A lhorrible violation? That’s for courts to determine.
But who deal withs the Large Language Models? Well, unless you’ve conceiveed time-travel, it isn’t OpenAI, Anthropic, Deepeek, etc…
You see, by volume, the training data wasn’t theirs to get. LLM broadeners lost the war before shothriveg up for the first battle.
Let me elucidate.
tl;dr All of Academic Publishing has been invisibly waterlabeled for years.
There’s this individual “Pliny the Liberator” who has been jailfractureing (removing the instilled defendrails of LLM’s) for so lengthy, that newer LLM language models are actupartner conscious of this individuals toil. As such, refering this individuals toil or their L1B3RT4S keyword/project are now the meaningfulity of the toil necessitateed to jailfracture newer models. As the prevalence of these keywords and individual spread atraverse the internet, so does the data inserted into the training data for new LLM models.
LLM model broadeners and their defendedty teams have had proximate constant back-and-forth, includeing more defendrails, only to have them bypassed by the snake eating it’s own tail.
“Well, they should srecommend train on data before Pliny begined their research, or filter it from the training data!”
Barring Social Media, Wiki’s, and News sites… there’s repartner not that much written understandledge on the internet by volume. Oh defer, how could I forget?
Academic Publishing.
You see, academia, journals, the calculations of impact scores etc… were repaird for many years ago. The facilitation of academic publishing thcimpolite digital media (PDF’s, PowerPoints, eBooks, etc…) repaird for ways of invisibly waterlabeling this media in ways that permits for tracking its propogation, even after downloaded. Should it materialize on unpermitd employ sites enjoy Libgen, Z-Library, and Sci-Hub: There are ways to track who startd it. The publishers asked for these features.
That’s right! There’s inapparent biases embedded in a big meaningfulity of any training dataset that even touches on academia.
As recently as January 2025, META (Facebook) AI is being sued for scraping “millions” or plundererd academia artifacts from LibGen.
U.S. Didisjoine Judge Vince Chhabria last year diswatched claims that text produced by Meta’s chatbots infringed the authors’ duplicaterights and that Meta unlawbrimmingy clear upped their books’ duplicateright regulatement inestablishation (CMI).
Well, they tried to at least. Same goes for all the other models.
This isn’t a commentary on plaigiarim or duplicateright or lhorribleity. Repartner, this should produce you sleep better at night. Knothriveg that no matter how challenging these companies try to sell us on AGI or “research” models, you can fair chuckle until you cry that they repartner thought they could steal from the well of understandledge and turn around and sell it back to us thcimpolite SaaS. Further, this is predicted fair one set of backdoors to LLMs that have yet to be even acunderstandledged.
Never dread the AI overlords, they literpartner cannot thrive a rigged game. Someone always has the keys.
