@morite2toeic As a native speaker: this is tricky and I had to think about it for a few seconds, but then I decided (D) is the only correct answer.
The [Laughs] is the key part. It means the previous comment was silly. Especially because the "Oh." means he got serious again.
@Rokugatunoaruhi@morite2toeic Respectfully, this is way off-base. As a native English speaker, "Yes please" and "Whiskey" have extremely low phonological similarity to my ear; I could never hear this "pun".
Mostly: "Please" is long and with emphasis, while "Whis-" gets the accent and "key" is short.
I was thinking about how we can look at query complexity relationships between multiple parameters.
First video: How our understanding of R/Q/C got filled out recently.
Second: Adding in D, there's exactly one missing vertex or plane.
Blog post:
https://t.co/bZJCj90xPx
The real problem with having the frontier models all run by giant companies instead of hosted locally is that when ChatGPT does something stupid, I want a button that says "that was so stupid, go do backprop RIGHT NOW and adjust your weights so it doesn't happen again"
@noahlt Because textbooks are written to be a teaching material, and encyclopedias are written to be a reference material. Wikipedia is pretty good as a reference material for me ("I know there's something like x, what is it exactly"), but bad for learning brand new things. By design.
@LinusMixson What language are the 650 lines in? Can make a world of difference. (Also, what kind of model faithfulness you're using.) Would be curious to see/try to help
@Lavader_ Extremely curious how killing people who hold 4% of wealth in the US is going to (checks notes) stop the military industrial complex, and raise teacher salaries
@blue_control69@netflix Yeah, is definitely an Entry-Point (Acquisition) project: get new people into MtG, get them to eventually buy packs. It's not Core or Fan-Service.
@ZachWeiner I would agree with all of the third column, if the row labels dropped the word "generally".
I think AI isn't "generally" good or bad, because it has distinct good and bad parts; any more than asking "Americans are generally white" / "generally nonwhite" / "have no inherent race"
@taktoa1@thinkingheights@thomasfbloom Yeah, I think that's a correct 'fix'. You can write "foo (fuel := 10^100) = true" as a term and Lean will reduce both sides to see that it evaluates the way you say it does. The problem is that a proof term can lead to lots of reduction steps.
@thinkingheights@thomasfbloom To be precise, let f(n) be any computable function (say, 10^10^10^n), and let g(n) be it's inverse (the smallest k such that f(k)≥n). There is a constant C such that every theorem can be proved in Lean with a term of size C + g(shortest proof of your theorem in ZFC, in steps).
@thinkingheights@thomasfbloom To be precise, let f(n) be any computable function (say, 10^10^10^n), and let g(n) be it's inverse (the smallest k such that f(k)≥n). There is a constant C such that every theorem can be proved in Lean with a term of size C + g(shortest proof of your theorem in ZFC, in steps).
@thinkingheights@thomasfbloom That's what I mean by "proof term size". There are, maybe surprisingly, ways this can still be hacked to be (essentially) constant size!