Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[silverspawn]

I also have this vague dejavu of having brought this up before. This is becoming a recurring problem.

[Awaclus]

I don't remember having this argument, but I probably made the point that I don't particularly care about being good relative to other people; I care about everyone being good relative to some objective level of skill. Put, I think being good at Dominion is a good meme.

If I made some other point instead, I may or may not agree with that point anymore.

[silverspawn]

Why care about everyone being good relative to an objective value?

[silverspawn]

Something I've wanted to do for a long time but have never gotten around to is to write a really in-depth article on why the series of unfortunate events is so interesting. I don't think there's any audience for it, but I still want to do it

[Awaclus]

Why care about everyone being good relative to an objective value?

Because I want to be the best version of myself possible and I want everyone else to also be the best version of myself possible.

[silverspawn]

Why care about everyone being good relative to an objective value?

Because I want to be the best version of myself possible and I want everyone else to also be the best version of myself possible.

[silverspawn]

Frequentism vs. Bayesianism is the only context in math where it's socially accepted to argue against a theorem. It's very strange. Everywhere else in math, it's understood that if someone proved something, that settles the debate. But in probability theory, where we have a straight-forward (provable from first principles in a few lines!) theorem, somehow it's okay to just go, nah I'm not doing it that way; I like frequentism better.

(I admit this is partially a strawman since not all frequentists legit do stuff that contradicts Bayes' theorem -- but some do, and the extent to which they do is still significant enough to justify the point.)

[silverspawn]

It's one of the many areas where problems arise due to the fact that the central conflict is ill-defined, so people can go around applying it in very stupid ways without getting called out since it's under the label "frequentist", which sometimes also belongs to only moderately stupid ideas.

[silverspawn]

The Incompleteness Theorem is a boring result about the capabilities for formal proof systems with zero philosophical implications.

[silverspawn]

^ one of my most contrarian takes that I genuinely think is true

[silverspawn]

This book INFURIATED me! What a lazy author. Taking us down the garden path through 12 books and then ending the series without really completing the story. As a children's librarian, I no longer recommend this series to kids. I hate to see them as disappointed as I was at this really stupid final book. Shame on you, Mr. Snicket!! You took the chicken's way out. Next time, have an idea of where you expect the series to go before you start it. Truly shameful.

Imagine thinking that *not* wrapping up everything with a neat little bow is the chicken's way out.

[silverspawn]

Anyway, I actually sat down and wrote a review for the series of unfortunate events! I figured better half-ass it a little than not do it at all!

[silverspawn]

Apparently there are various other works from the author accompanying the series. I'm very skeptical that they're any good. The idea off adding things to the series is almost inherently broken. and I feel like the author already kind of sold out by giving the okay for the netflix series to throw all the most central theme of the book in favor of an asinine happy ending/conclusion.

[silverspawn]

I wanna punch everyone who called the last book lazy

[faust]

The Incompleteness Theorem is a boring result about the capabilities for formal proof systems with zero philosophical implications.

I resent the implication that results about capabilities of formal proof systems are boring.

And it's a weird take for someone interested in AI too. Formal logic is an important piece of the puzzle in developing systems beyond the truly boring neural network stuff.

[silverspawn]

Yeah, once upon a time, before I cared about AI, some people seemed to have thought AI systems would be like little agents solving theorems in their own formal systems. But that's so far away from what anyone does right now that I don't see it coming back

[silverspawn]

So the three largest poker sites right now are

GGPoker
PokerStars
PartyPoker

Last time I checked, PartyPoker had a rule specifically for Germans (just for me, yay!) that made you pay an additional ~20% on top of rake as an entry to every game, to... discourage gambling? It was very strange. I don't know if there's any player in the world good enough to make a profit under that rule. So that's that. Afaik, no other site does that.

GGPoker is the biggest $$$$ pile of $$$$ with $$$$-$$$$-$$$$ interface $$$$$$$$. It starts off not letting you choose where to install, then it went into a very slow installation with a progress bar that took longer than that of any other client, then when it was done (I was going to go to sleep after installing) it started another progress bar, and I have no idea how many it was in total, but perhaps 3* the length of the original one. None of them cared to inform me what was happening mind you. Then when you get there, it blasts obnoxious Christmas music without asking you first and has a ridiculously cluttered and obnoxious interface. It's like they were optimizing for me to hate it. It's also now the largest poker site. It has no sit&go's (????) (these are small on-demand tournaments and my default way to play) but does have cash games, hidden somewhere between great alternative formats like ALL-IN OR FOLD.

PokerStars is a mildly annoying, reasonable tasteful interface with very good customization. Also has some nonsense gambling formats and blackjack or whatever mixed in.

So the upshot of all that is that... if I ever play cash games, I should probably play them on GGPOker? The interface is so obnoxious that I just cannot believe there isn't at least a small positive correlation between people who prefer PokerStars and their skill. So you'd want to go where the weaker players are. And the rake is slightly lower as well.

[silverspawn]

At least they got rid of the thing where there's the 24 hour lockout button which is at the top of every table, and if you press it once by accident, it irreversibly locks you out for 24 hours. The button still exists, but now it's not on every table anymore, and it says that you have to click and hold. There's also a warning after every 60 minutes that you're playing for too long. This is the level of annoyance that I think makes sense to discourage gambling.

The button was so stupid before, I think I've even written about it then

[silverspawn]

Speaking of roulette*, there was this super interesting comment in the history-of-poker video from the MIT lecture that I linked somewhere on the previous page. He was talking about how he was hanging out with a bunch of smart mathy people, and they were talking about whether roulette is abusable. Someone said it depends on whether the wheel has regular or irregular behavior when it's spun, but someone else said that it will be breakable either way. If it's regular, then you can compute where it lands, so you can abuse it. But if it's irregular, then that means there's an uneven distribution of where it could land, which means well that you can abuse it statistically. You only need to do slightly better than random to make a profit in blackjack.

Apparently the way it does work is that it starts off regular until the very end where the ball spins rapidly in the middle, but that process is irregular. So if you compute where it starts entering the irregular phase, you get a skewed distribution of where the ball lands. (I didn't understand how exactly you're supposed to compute the first part in real time since I think it depends on how the instructor spins it? But well I'm about to try). Anyway, super interesting stuff.



The dichotomy is not absolute; you could have a regular process that is just too hard to compute. But apparently, we don't know how to make devices like that yet.

Is computer RNG good enough? I always thought so, but I never actually researched it.

*I know I wasn't speaking of roulette at all, but I confused it with blackjack in my head and then I was already writing the comment, so 

[silverspawn]

The Incompleteness Theorem is a boring result about the capabilities for formal proof systems with zero philosophical implications.

I resent the implication that results about capabilities of formal proof systems are boring.

And it's a weird take for someone interested in AI too. Formal logic is an important piece of the puzzle in developing systems beyond the truly boring neural network stuff.

btw yes, I agree, it's mathematically exciting just for what it is. But like, it should be exciting for mathematicians only.

[faust]

Yeah, once upon a time, before I cared about AI, some people seemed to have thought AI systems would be like little agents solving theorems in their own formal systems. But that's so far away from what anyone does right now that I don't see it coming back

We'll see. The neural network hype is beginning to die down. I am of course biased since I am indirectly associated with several people working in the field of automated proof systems.

[silverspawn]

I do agree that the neural network hype has begun to decrease. But it's still a really high levels. I feel like it'll have to decrease for several more years before automated proof systems catch up in hype, although I haven't done a literature analysis or anything.

It would be pretty funny from my perspective if it did in fact happen. The Yudkowsky-ian view on AI has been criticized a lot recently for having mis-diagnosed the ways AI would come to be -- Eliezer has been on record saying that the neural network approach seems silly, and that you have to actually understand what you are doing. A lot of the work Miri has done, if not all of it, makes a lot more sense if AIs are proof systems than if they are LLMs. However, Eliezer has not maintained this view throughout the LLM hype; he's very much changed his mind somewhere between GPT-2 and now, and adopted very short timelines. So I guess whatever happens, he will have been wrong at least once.

I'm still more worried about people figuring something out by combining LLMs with some kind of other system in a way that makes it do iterative reasoning, than I am about proof systems. The reason being that LLMs have been much more impressive than proof systems so far afaik. Although my median prediction is that neither of them works and there will be a paradigm shift toward non-digital and conscious computing.

[silverspawn]

Speaking of both approaches, I don't really see them being compatible. Proof systems are highly formal and LLMs highly non-formal; the task of translating LLM output into anything usable by a proof system seems extremely hard; too hard to be useful.

[silverspawn]

No it's okay GGPoker, I didn't want to know where you're installed anyway. I just checked all the regular places for fun. It's completely fine that even your desktop shortcut which you created without my permission just links into some BS roaming/appdata directory that doesn't answer my question either.

[silverspawn]

The important thing is that you have cool animated snow on in the interface and lots of flashy advertisements. Gotta have priorities.