Nah, I think GP is spot on. I took a quick look at some of the materials and I would say that they're pretty outdated as far as current understanding of the game goes. It's more in line with how people talked about & thought about the game in the boom era (the 00s) or possibly even earlier, to be honest.
If I were teaching poker in a university setting I would generally work with a bunch of toy games to teach concepts of polarization, MDF, indifference, and so on. Those are the fundamentals of poker theory in the modern environment.
I'll give you a concrete example. In lecture five at 15m he starts talking about donk betting, which is when you lead out with a bet on a street into the aggressor from the previous street instead of checking to them to let them bet (for example, you're the big blind, you call a raise from the button preflop, and then when the flop comes you bet directly yourself instead of checking). He mentions that this is unusual, which is true, because normally you would just check and let the person with the betting lead bet into you if you have a good hand, and so the donk bet is typically weak--but then he goes on to say that advanced players may exploit this perception by donk betting with a strong hand knowing you will interpret it as weak, and then you can exploit that line of thinking, etc.
A more modern view of donk betting is this: the "betting lead" is not anything inherent to the game but just an artifact of the range of hands each person has. Typically when you call preflop from the big blind, you'll have a wider range of hands than the person raising preflop, because they can't raise too wide profitably due to players left to act after them, and you get to both close the action & also get better pot odds on your call from the big blind. Not only do you have a wider range of hands, then, but you also lack the strongest hands (AA, KK etc) because you would have reraised (3bet) them preflop. So overall, on most flops, the preflop raiser will have a stronger range of hands and can thus be expected to "take the lead" betting.
But! There are some flops that can neutralize or even reverse this range advantage. The flop 654 rainbow is now better for the big blind than for an early position (EP) raiser, because the big blind will have many straights, sets, two pairs, and pair+draw hands in their range, while the EP player will generally not have these hands. The EP's big pairs (AA, KK etc) are less strong, and some of their other strong hands preflop like AK or AQs have totally whiffed too.
As a consequence, it is correct for the big blind to have a donk betting strategy in this situation--the way the two ranges interact, the big blind is now incentivized to put money into the pot directly, and in fact, if the big blind does not have a donk betting strategy on this flop, the EP player should respond by virtually never betting themselves. The big blind is incentivized to bet frequently with a small sizing and put pressure on EP's overcards; if you have, say, 86s, and you can get a hand like QJo to fold, that's a pretty decent result on the flop, because QJo has six outs to improve to a better pair.
Anyway, sorry that this example was probably hard to follow if you don't play poker, but it's probably illustrative of some of the ways in which poker theory has evolved over the years--more focus on the specifics of range vs range interactions. There are many more complicated and intricate examples on turns and rivers that solvers are very good at finding but may or may not be obvious to humans.
> It's more in line with how people talked about & thought about the game in the boom era (the 00s) or possibly even earlier
Note that most poker games were very easy in those days. You'd log in and enter a couple of tables where most of your opponents were going to take the name 'Hold'em' very literally, also some of them were drunk and you could basically expect them to bet with anything without paying any attention to your table image.
So the optimal strategy in those days was to simply wait for solid hands, you would get paid for them anyway even if you had been folding everything for the last hour.
Good detailed example. One simple one I like to use without going too much in depth - it used to be considered heresy to simply complete in the small blind when the action had folded to you - you were supposed to raise or fold and if you deviated from that in the early 00’s you’d have been widely mocked.
Solvers love completing the sb. They also occasionally like to open limp on the button against certain strategies - another thing that would have been sacrilege not that long ago that’s completely commonplace now. The “donk bet” once belonged there too.
Have you guys considered that maybe he's there because he's extremely qualified and extremely well-respected by his peers? It's not some kind of weird power play, he's just lending a hand while they figure out the long-term board composition.
Adam was appointed to the OpenAI board in April 2018, long before ChatGPT and Poe. He's always been somewhat interested/involved in AI/ML so the appointment broadly makes sense to me.
Also keep in mind that a year earlier in Spring 2017 Sam Altman led Quora's Series D, after YC previously joined in on Quora's Series C in 2014. So the two of them clearly had some pre-existing relationship.
I don't think OpenAI and Quora (the product) are a serious conflict of interest. You claim "I'm sure Quora views took a hit after ChatGPT" but I really doubt that's true in any meaningful way. Quora's struggles are a separate issue and predate the GPT craze of the last year.
Nor were Poe and OpenAI competitors until recently; Poe was simply building on top of OpenAI models, the same as hundreds of other ventures in the space right now.
However...I do agree that the GPTs announcement two weeks ago now creates a very clear conflict of interest--OpenAI is now competing directly against Poe. And because of that, I agree that Adam probably should leave the board.
The timing also raises the question of whether booting Sam is in any way related to the GPTs launch and to Poe. Perhaps Sam wasn't candid about the fact that they were about to be competing with Adam's company. The whole thing is messy and not a good look and exactly why you try to avoid these conflicts of interest to begin with.
I never said Adam should've never been on board. I was arguing about the part after Poe was competing with OpenAI after DevDay. That's where he has a clear, very strong conflict of interest and to be honest that's where the board/Adam took the most impactful decision that OpenAI board ever made.
On the other hand, Sam & Greg had the opportunity to confront Adam about the obvious conflict and likely could have forced him to step down if they wanted him to. They made their choice. Zero mention about Adam & Poe in the leaks from Sam's camp also suggests Sam doesn't fault Adam's character here.
I didn't read OpenAI's company charter, but forcing Adam down would probably require a board majority. It's not like they would have made Adam step down if they wanted to.
Nope, Poe was always building on OpenAI API and their GPTs. In fact Poe was one of the first companies to get access to GPT-4-32k context length a few months ago and they were the first to make it accessible to their users.
As all Geoguessr players know, official Street View coverage in Germany was extremely limited and all very old. There was a small amount of coverage in around a dozen or so of the largest cities that dated back to the mid-2000s and that's about it. There's also various bits of crowdsourced unofficial coverage in the country which you are perhaps referring to, but it's very limited. Just go to Google Maps, look at Europe, and grab the Street View person, and you'll see.
The Twitter thread you linked is just somebody pointing out poor performance in Tigrinya. That doesn't mean it performs poorly on all non-English languages. It's performing poorly on the languages with little representation in the corpus, but those are also the languages that are least likely to be learned on Duolingo. Duolingo users mainly learn languages like English, Spanish, French, German, etc. and the models do just fine with these languages.
If I were teaching poker in a university setting I would generally work with a bunch of toy games to teach concepts of polarization, MDF, indifference, and so on. Those are the fundamentals of poker theory in the modern environment.
I'll give you a concrete example. In lecture five at 15m he starts talking about donk betting, which is when you lead out with a bet on a street into the aggressor from the previous street instead of checking to them to let them bet (for example, you're the big blind, you call a raise from the button preflop, and then when the flop comes you bet directly yourself instead of checking). He mentions that this is unusual, which is true, because normally you would just check and let the person with the betting lead bet into you if you have a good hand, and so the donk bet is typically weak--but then he goes on to say that advanced players may exploit this perception by donk betting with a strong hand knowing you will interpret it as weak, and then you can exploit that line of thinking, etc.
A more modern view of donk betting is this: the "betting lead" is not anything inherent to the game but just an artifact of the range of hands each person has. Typically when you call preflop from the big blind, you'll have a wider range of hands than the person raising preflop, because they can't raise too wide profitably due to players left to act after them, and you get to both close the action & also get better pot odds on your call from the big blind. Not only do you have a wider range of hands, then, but you also lack the strongest hands (AA, KK etc) because you would have reraised (3bet) them preflop. So overall, on most flops, the preflop raiser will have a stronger range of hands and can thus be expected to "take the lead" betting.
But! There are some flops that can neutralize or even reverse this range advantage. The flop 654 rainbow is now better for the big blind than for an early position (EP) raiser, because the big blind will have many straights, sets, two pairs, and pair+draw hands in their range, while the EP player will generally not have these hands. The EP's big pairs (AA, KK etc) are less strong, and some of their other strong hands preflop like AK or AQs have totally whiffed too.
As a consequence, it is correct for the big blind to have a donk betting strategy in this situation--the way the two ranges interact, the big blind is now incentivized to put money into the pot directly, and in fact, if the big blind does not have a donk betting strategy on this flop, the EP player should respond by virtually never betting themselves. The big blind is incentivized to bet frequently with a small sizing and put pressure on EP's overcards; if you have, say, 86s, and you can get a hand like QJo to fold, that's a pretty decent result on the flop, because QJo has six outs to improve to a better pair.
Anyway, sorry that this example was probably hard to follow if you don't play poker, but it's probably illustrative of some of the ways in which poker theory has evolved over the years--more focus on the specifics of range vs range interactions. There are many more complicated and intricate examples on turns and rivers that solvers are very good at finding but may or may not be obvious to humans.
source: semi-professional online cash game player