If by plagiarism you mean that the LLM passes around content protected by copyright then no. If you read a book and then talk about it with someone else are you plagiarizing? Are book reviews plagiarism? The LLM is like a matrix of weight distributed word associations created by running the entire internet (including pre-internet content thanks to OCR) through the machine. It doesn't return you pre-calculated responses.
Thanks for your comment. I read your blog post, and I agree with you that even language like "hallucinates" sort of obscures the machinery of what's happening (a malfunctioning brain? Really?). We do this andromorphic exercise with tech generally--ships becomes "she" somehow to Navy admirals and captains etc. But the language is sort of endearing, or it serves some pragmatic end, it's not potentially fundamentally confusing, so I think the cases are in fact different. Thanks for your comment again. It's gratifying to see folks engaging the material. My hope is that we can actually solve or help solve some of these problems one day. (Big hope.) Best, Erik
This is great. Thanks for the work. I wrote a piece on my own blog (https://www.keithlowery.com/is-ai-demonic/) a while back trying to explain what language models do and what we might want to have some concerns about. The short answer is that there is nothing wrong with the models (other than correctness), but there's something very wrong with the effort to stampede everyone into viewing them with veneration and awe. Your comments about the effort to conceive of these models as "a near religious artifact" was reminiscent of my own concerns. Also your point about how the models are distilling (and profiting from) content they don't actually own raises all kinds of copyright questions that are far from resolved. I share your concerned. After my first experience with ChatGPT I raised the similar question of whether it might not be, essentially, just a giant, computationally rigorous plagiarist. (https://www.keithlowery.com/the-granularity-of-plagiarism/) I'm enjoying Colligo. Thanks for doing it.
Persuading the general public that narrow AI can make “more sound and eloquent decisions” to enable outsourcing of decision making functions in all sectors to “ai” systems. Decision making about such fundamental questions such as personal well-being. A well known and time tested truth “heathy body healthy mind” is now being replaced with 10,000 steps and the erosion of self discipline in eating.
Surely there are some ambitious IP attorneys who would be willing to file a class action suit against this egregious copyright violation on behalf of humanity!
Aren't the large language models just mega-plagiarism machines?
If by plagiarism you mean that the LLM passes around content protected by copyright then no. If you read a book and then talk about it with someone else are you plagiarizing? Are book reviews plagiarism? The LLM is like a matrix of weight distributed word associations created by running the entire internet (including pre-internet content thanks to OCR) through the machine. It doesn't return you pre-calculated responses.
Hi Keith,
Thanks for your comment. I read your blog post, and I agree with you that even language like "hallucinates" sort of obscures the machinery of what's happening (a malfunctioning brain? Really?). We do this andromorphic exercise with tech generally--ships becomes "she" somehow to Navy admirals and captains etc. But the language is sort of endearing, or it serves some pragmatic end, it's not potentially fundamentally confusing, so I think the cases are in fact different. Thanks for your comment again. It's gratifying to see folks engaging the material. My hope is that we can actually solve or help solve some of these problems one day. (Big hope.) Best, Erik
This is great. Thanks for the work. I wrote a piece on my own blog (https://www.keithlowery.com/is-ai-demonic/) a while back trying to explain what language models do and what we might want to have some concerns about. The short answer is that there is nothing wrong with the models (other than correctness), but there's something very wrong with the effort to stampede everyone into viewing them with veneration and awe. Your comments about the effort to conceive of these models as "a near religious artifact" was reminiscent of my own concerns. Also your point about how the models are distilling (and profiting from) content they don't actually own raises all kinds of copyright questions that are far from resolved. I share your concerned. After my first experience with ChatGPT I raised the similar question of whether it might not be, essentially, just a giant, computationally rigorous plagiarist. (https://www.keithlowery.com/the-granularity-of-plagiarism/) I'm enjoying Colligo. Thanks for doing it.
Persuading the general public that narrow AI can make “more sound and eloquent decisions” to enable outsourcing of decision making functions in all sectors to “ai” systems. Decision making about such fundamental questions such as personal well-being. A well known and time tested truth “heathy body healthy mind” is now being replaced with 10,000 steps and the erosion of self discipline in eating.
Surely there are some ambitious IP attorneys who would be willing to file a class action suit against this egregious copyright violation on behalf of humanity!
A win-win! /sarc