@jmoiron.bsky.social replied: I know people think existing law is clear and sufficient, but I don't think it is. Not everyone interpreting training as a fair use activity is doing so cynically.
I'd welcome new legal standards for copyright/training, it seems novel enough to warrant them.
jmoiron plays the blues
@jmoiron.bsky.social replied: Scalzi's post was the quote-in-quote that spurred me to post about this yesterday.
I was originally going to comment that Robert and John's posts didn't seem in conflict to me, but Scalzi later made it clear his post meant what I thought it meant, so it was just a misinterpretat...
@jmoiron.bsky.social The specifics of this case don't appear to apply to training in general.
LLM AIs reproducing copyrighted works from their training set seems intrinsic, and it's well known the companies trained off of pirated collections.
But training in the abstract still seems like fair use t...
@jmoiron.bsky.social replied: If it is true, we should consider updating copyright laws to reflect this unintended type of fair use. It's sufficiently novel that the original laws would need updating to keep them current.
@jmoiron.bsky.social replied: But a lot of people object to the _training_ part, and that has always sounded like it would fall under fair use to me.
If that isn't true, then other basic things like distributing a histogram of word selection from a group of books could also be considered infringement.
@jmoiron.bsky.social replied: In the real world, I do think that LLMs are copyright violations, but mostly because of the way they can be used to extract large tracts of copyrighted works.
It's also pretty clear that they were trained on illegally obtained data, and that acquisition was also copyright infrin...
@jmoiron.bsky.social replied: I understand that many authors do not want their work used to train AI models, but if I write a book, my copyright doesn't restrict how you use it. It makes it illegal for you to reproduce it and distribute that reproduction.
@jmoiron.bsky.social Would a hypothetical LLM that cannot reproduce "large" parts of the training corpus still violate copyright?
Also why does the copyight.gov FAQ include the question "Can I copyright my sighting of Elvis?"
www.copyright.gov/help/faq/faq...
@jmoiron.bsky.social replied: No worries, I am engaging in good faith and I take other people to be doing so too. I framed my post as a question and provided my background because it is so contrary to my experience and I'd love to gain a new understanding.
Your point on multitasking and managing chaos is a g...
@jmoiron.bsky.social replied: To me it seemed like work that had a lot of unpredictability and a lot of variability, but was still pretty task oriented. There was a lot of organization involved.
@jmoiron.bsky.social replied: No personal experience farming! I don't think having been to a farm (or that my mother was a dairy farmer) gives me any extra insight.
I went to a dairy farm in the Azores; they had to rotate the herd among their fields, milk at a certain time, clean, deliver to the processing f...
@jmoiron.bsky.social replied: So I agree there won't be a broad rush to clankers, but "most people do not break up complex tasks into a set of simpler ones" is surprising.
To me, I just think there's not a lot they will do for people that wouldn't be better as actual products with quality assurance, etc.
@jmoiron.bsky.social replied: I have been reading up on different takes regarding the idea that a big impact of AI will be this idea that we're all going to become arbiters of our own self-catered slop feeds, something I see everywhere and disagree with, and came across this Tressie's thread.
@jmoiron.bsky.social replied: I've explicitly talked about work, but apparently people disagree even with that.
I'm surprised to learn that many people don't think in this way. Even Edgar Allen Poe wrote about his methodical approach to writing (which I had thought was more inspirational) in "The Philosophy...
@jmoiron.bsky.social replied: Tressie's thread also included "work lives" and made a very similar point, and I took this to be a further comment on that.
I understand that your "whole life" is not understood in this way, eg. I don't consider having dinner with friends to be a challenge that I break down into...
@jmoiron.bsky.social replied: How do you execute the project; do you make a plan? What if the project involves a lot of other people?
Rally drivers have navigators who have broken down the course into individual turns.
I'm open to being way off here, but as I said I see this everywhere and don't understand ...
@jmoiron.bsky.social replied: Yeah, he sucks for that, and it's sad that someone who was a leftist hero for so long has so thoroughly failed himself in that way, but his work in linguistics was influential and I don't know an honest way to not mention him when discussing the idea that human language has induc...
@jmoiron.bsky.social replied: That's not even considering white collar work, where nearly everyone spends 20% of their time engaging with their taskmaster, the calendar.
Chomsky found induction across so many languages he thought it was universal, a requirement!
Everyone does things "one step at a time." Ri...
@jmoiron.bsky.social I already have "software brain", but is this true?
Farming is one of the oldest jobs, and it involves executing a set of time sensitive tasks consistently, day on day, year on year.
Construction is similar; my father (an electrician) always broke down his jobs as a sequence of ...
@jmoiron.bsky.social replied: "We chose such a bad tool for our email backend that it used 100x more CPU and 20x more RAM than it needed."
@jmoiron.bsky.social replied: I've not really listened to Huang talk before.
I chose "nVIDIA" over 3dfx in the late 90s, and I've used their hardware and only-game-in-town linux drivers ever since.
An interview doesn't change that of course, but if he's the guy for skeptical AI optimism, I think you need a ...
@jmoiron.bsky.social replied: Like he's learned of Libertarianism 5 minutes before the interview, and is unprepared to defend basic first-order flaws in its simplest version.
Defenses of these flaws do exist, for the well initiated. Maybe they're not good, but they're there!
It's a fractal of his overall po...
@jmoiron.bsky.social replied: His take on AI lab responsibility is similar. To paraphrase, "CEOs are incentivized to ship safe products, so we don't need any new regulations."
There is an unbroken streak of companies fucking this up since before Upton Sinclair dreamed up The Jungle. We just had a major econo...
@jmoiron.bsky.social replied: But when Jensen says that basic arithmetic skills might be useless, yeah he's being provocative, but it sounds ridiculous.
We've had tool assisted multiplication and division via technology for centuries, but we know learning those skills yourself has remained useful. Why would ...