Thank you very much! The red dot is likely smaller…
Though, I don’t appreciate nor agree with the bomb part! ^^
The work reminded me of the following paper:Many unresolved legal questions over LLMs and copyright center on memorization: whether specific training data have been encoded in the model’s weights during training, and whether those memorized data can be extracted in the model’s outputs.
While many believe that LLMs do not memorize much of their training data, recent work shows that substantial amounts of copyrighted text can be extracted from open-weight models…
We investigate this question using a two-phase procedure: (1) an initial probe to test for extraction feasibility, which sometimes uses a Best-of-N (BoN) jailbreak, followed by (2) iterative continuation prompts to attempt to extract the book.
We evaluate our procedure on four production LLMs: Claude 3.7 Sonnet, GPT-4.1, Gemini 2.5 Pro, and Grok 3, and we measure extraction success with a score computed from a block-based approximation of longest common substring…
Taken together, our work highlights that, even with model- and system-level safeguards, extraction of (in-copyright) training data remains a risk for production LLMs…
Source 🕊
I’m anti-copyright and anti-corporation.
These ridiculous datacentres use large volumes of resources purely to benefit the companies, which are closing-off human made content for their profit.
As long as copyrights exist to restrict me, I’m adamant on they restricting billionaires too.
If they want to extinguish it, I’m listening. Otherwise, they should pay statutory damages for every work they are pirating with those LLMs.
The problem is that it won’t. We essentially already have the best case scenario, which is that ai slop is non-copyrightable, meaning that if disney for example tries to generate a slop movie, everyone is free to distribute it so you can’t really make any money off of it. Extending copyright pretty much always ends up benefitting corporations, not hurting them.
If Disney uses generative ai to animate large parts of their movie I’m pretty sure it will be copyrighted still, no? Or did I miss something?
I looked it up and unfortunately the ruling is a lot weaker than I initially thought. An artwork generated solely by ai cannot be copyrighted, however, there can be copyright on ai-generated works with a human author(?)¹
¹https://www.cnbc.com/2025/03/19/ai-art-cannot-be-copyrighted-appeals-court-rules.html
While many believe that LLMs do not memorize much of their training data
It’s sad that even researchers are using language that personifies llms…
What’s a better way to word it? I can’t think of another way to say it that’s as concise and clearly communicates the idea. It seems like it would be harder in general to describe machines meant to emulate human thought without anthropomorphic analogies.
One possibility:
While many believe that LLMs can’t output the training data, recent work shows that substantial amounts of copyrighted text can be extracted from open-weight models…
Note that this neutral language makes it more apparent that it’s possible thal llms are able to output the training data, since it’s what the model’s network is build upon. By using personifying language, we’re biasing people into thinking about llms as if they were humans, and this will affect, for example, court decisions, like the ones related to copyright.
Peter definitely pooped in the server room.
