I mean, according to the graph he's losing money, so not that great I guess? :)

It's a fairly meaningless stat without knowing the number of the players at the table. At a quick glance he seems to be playing 6-max, but sometimes 3-handed. In any case 40% is within the reasonable range for 6-max.

Oversimplifying for sure but if you're loose and aggressive against a tight aggressive player, you're going to make them fold most of the time and win a small amount by applying pressure on bluffs but every once in a while if you get too aggressive and they call you because they have a monster hand then you get wrecked

No idea, obviously you can see I'm a n00b and not a profitable poker player but on the strategy side, and this is very oversimplified, but you should be folding a lot of the time other than when you get AA-22, AK, AQ, AJ, AT, KQ, KJ, QJ, JTs, T9s, 87s you call or 3 bet pre-flop because you have good odds. When you're up against tight players you can make a small bluff on the flop and scare them away most of the time, if they raise you fold though. Position is very important in the game, when you're on the button you have odds in your favor because everyone else has to check or bet before you so you play more loose and aggressive in that position and more tight and passive in early position. There is no one single strategy to memorize and apply, that's why it's great. 5 minutes to learn the game, a lifetime to master.

The "standard" strategy is to play GTO (game theory optimal). There are solvers out there (like GTO Wizard) that show you the "optimal" play for every situation, which is used as a baseline, and then players deviate to exploit specific player tendencies.

GTO trees are far too complex to fully memorize, so nobody can play perfectly GTO. But you can do a lot of solver work to get reasonably close.

And the best hackers at any level abstraction will always be the ones that actually understand what's going on in every lower layer in order to diagnose when the abstraction is failing them. Anyone that thinks you can be up at the level of vibes without understanding how an LLM thinks, without knowing how to review and factor your vibed Python or whatever high-ish level language, can make it performant without knowing if or when to write something in lower levels like C or need to be using a library where all the hard work is in something like C, make it secure without understanding how that gets turned into instructions for an incredibly obedient but ignorant machine (like the LLM is but in the exact opposite ways, buffer overruns and free before use and stuff)... It's a holistic practice. The guys that produce code and don't know which parts are happening in the browser or in the client, think they can trust the values of cookies not to be tampered with and junk are able to be productive, probably more so with LLMs these days but they simply can't make quality software and never will be able to. Corporations love them because nobody's accountable to resiliency (securty, quality, reliability) until something actually breaks and those guys can get thrown under busses easily when and if that happens because they're cheap cogs. Hackers love them because we'll always have work to do to improve (or compete with, or exploit) what folks like that make.

Good take

Exactly. This is how it's always been. LLMs make it easy to spit out boilerplate code, which drives the price of boilerplate down to free. But good engineers will add a lot more value to that which raises the bar for everyone. The things you can create with an LLM become boring and worthless (honestly they mostly already were before coding agents came out) and the hirable skills become everything else that engineers need to do.

Given how awful LLM are at chess, I'd say GPT sucks at poker. Making a profitable bot using state of the art poker software, like stockfish for chess? That's already done.