xkcd 2347, but edited by me to say "All post-WWII popular culture" for the mass of things built on the shaky foundation and "Lax enforcement of copyright law" for the critical single piece of that foundation that, if removed, would topple the whole thing
what I wish I could scream whenever I see artists clamouring to extend the scope of copyright and enforce it harder, be it to stop AI or for any other reason
A tool that can be used to break boundaries in math cannot be called a plagarism machine when it has nothing to copy from. That we're seeing results in a field as fundamental as maths, from the greatest minds alive, that tells me something.
We're just figuring out how to use it still. Wild shit.
the post proves its point; people are reacting to this as if it’s some outrageous claim instead of just a literally accurate description of how large deep learning models work, one repeated by every researcher and scientist, including both the 2024 chemistry and physics nobel winners
Similarly to how nothing makes me more hostile to AI-boosterism than some of the dumber breathless hype you see on X/Linkedin, nothing makes me hostile to AI-skepticism than the sneering confident incuriosity that you see in stuff like the quotes of this. This is a totally valid metaphor!
Pangram iteration?
Internet artists think that IP law binds corporations and prevents them from using their work, while also freely permitting small-time non-profit fanworks, or works for minimal profit.
These are the people who have frequently told me I don't understand copyright law.
Small optimizations that could be made:
ReLU instead of GeLU for activations
RMSnorm instead of LayerNorm
The weight dimensions should be ordered to make the matrix products Cracovian
I meant like let that occur internally, and have the software float type be a FiniteFloat, while a tagged union of that, possible infs, and possible nans correspond internally to the full ieee float domain
It's the cheapest thing that is Known To Work. What more would you want?
IMO having infs and NaNs be lumped together in the same type as the finite float values is a mistake of similar kind to null pointers, but it's one that almost all programming languages have so I'm not harsh about it
So let’s make the laws even more draconian and create thousands more Swartzes
you might reasonably wonder why software engineers have so quickly embraced tooling that threatens to replace them. maybe programmers have good intuitions about task creation bc they see the tools in action and immediately think of 10 things to do that were too tedious to bother with before
I'm kinda being constantly tortured online bc I both think
1. It's extremely unlikely current AI models are conscious.
2. Almost every argument I scroll by for my side is insane and is just steamrolling decades of philosophy of mind work.
I choose to believe people are either willfully or spifefully misinterpreting that "Are humans conscious" post, because the alternative is y'all bitches are just really that dumb. You can't define consciousness, and thus you cannot definitively prove its presence *or* its lack!
We’ve been handed a really interesting philosophical question on a platter and it’s a bummer how much of trh reaction to it amounts to sticking your fingers in your ears and screaming
I lean very strongly towards “LLMs are not conscious” but I think they really demonstrate how we don’t understand the consciousness other than our own and the replies to this post seem to focus on a “god of the gaps” philosophy in which anything a machine can do is not a sign of consciousness
First question: what event space is being conditioned over here
If you're too dense to grasp this, the problem with that bill that aims to regulate "legal advice" and "medical advice" from LLMs is those are BROAD CATEGORIES and if politicians are able to make "being trans in public" equal pornography, imagine what they can to do with a ban on "medical advice".
Meme image of obama awarding obama a medal
People that evaluate themselves on knowing their own cognitive processes:
any time the government tries to restrict my access to information i think they're bad and wrong.
lots of people ostensibly agree with this until it's about a regulation from the left, then they don't. i actually believe this and the burden for overcoming it is unfathomably high
I put a copy of my darling into a .tar.zstd in backup storage first
In many cases, the models themselves are the most compute intensive thing running on the system
additional nail in the coffin of model collapse: better results so far on a model retrained on its own synthetic traces.
We should really get people comfortable with a rapid release cadence for history if this is the reaction to an update after multiple decades
Synthesized
Usually
It's the exceptions that can be scary
It's hard to convey how nuts it feels when just about every few weeks to a month tops, some perfectly acceptable research establishes LLMs can do mental reasoning/metareasoning tasks critics say are impossible, and not only does it fail to register - it leads to increasingly aggressive dismissals
No, the US Supreme Court did not declare that AI works cannot be copyrighted
If you have been online recently you may have been a variation of this story that reads something like this: "The US Supreme Court declares that AI generated works aren't copyrightable." I won't name and shame all of the…
I have had people dig through my social media and pluck out a silly joke I made about being a human grappling with technology and try to use it as an example of why I can't do my job and shouldn't be paid. I have had that happen multiple times. I am not patient with this kind of behavior
