Thank you for links.
I don't disagree with how much bot can improve, but if you want to draw analogy to chess. EV or WR is more analogous to evaluation bar in chess and if you watch GM play vs stock fish it's not like stock fish is +2 after 5 moves. Sure stock fish will win every game but it won't mate him in 10 moves.
I see 30bb/100 more like mate FM/GM in first 10 moves and you can't crush good player like that in poker or chess.

I think thats a great way of looking at it. I will certainly not be surprised if I am incorrect on this matter, but I want to explain more why I feel like this.

So I have been mentioning Chess a lot, but I have also personally observed a lot in poker. My memory is gonna be a bit hazy for events years ago so I'm also looking at old threads as I write this post too. But I started being very involved in poker in 2012, and it's basically gone like this:

~2012-2015: Solvers are not really a thing at all (I obviously cannot say whether some players were using secret cutting edge tech, but definitely most players were not: check out RedBaron's thought processes in his blog on 2+2 for example). It's crazy for me to think about but as recently as 6 years ago, people were regularly asking "What percent should I raise from this position?" and no one REALLY knew, it was often answered by "oh Red Baron plays this much VPIP/PFR, Sauce plays this much, etc. so somewhere between that is probably good". Upswing's preflop ranges in 2015 were top tier for the public and they basically made them by having a bunch of good players get together and go "hey I think raising 14% preflop in this spot is good, which hands are those oh Great and Powerful Poker AI Equilab?" (OK I don't know if that is true for Upswing specifically but definitely some HS crushers were). Also many people used preflop ranges from "The Course" was published this in 2015.

2016: Pio was released in September.

2017: Solver adoption spreads slowly throughout the poker community. But here's the thing, solved preflop ranges weren't readily available so even the outputs would always be much fuzzier than they are now (and again I could be wrong but I believe that 5 years from now, we're gonna think our 2021 ranges are fuzzy as ****). Lots of people reject them outright (and some go in the opposite direction and treat it as gospel). Many people said "you don't ever need to look at a solver if you aren't playing highstakes".

2018: Notably for PLO players, JNandez launchesd PLO Mastermind in 2018 which started slowly changing the landscape of PLO. This basically brought solver results to the public for PLO in an easily digestable manner, but very few people are actually working with the solvers themselves.

2019+: Even many of those who do actually start studying solver strategies are misusing them. I saw plenty of twitch streamers running NLHE sims in 2019 without even allowing overbets on any street. I watched Landon Tice in 2020 fiddle with various inputs over and over and re-running them until his poor solver was like "ok ok man this combo is only folded 80% and checkraise bluffed at 20% frequency, can I rest now" and he was like "alright my play was OK there, it was GTO approved". Not trying to shade Landon, I think he is one of the more intelligent/diligent winning players and was also in a circle of elite poker players- my point is, it could happen to anyone.

If you want an amusing trip down memory lane, look at the GTO discussion in this 2011 thread about Dwan vs GTO Bot. I mean a ton of the comments are literally asking what GTO stands for.

At any rate, we've come a long way in 10 years. I fully understand I kinda derailed from the original point, and that the solvers themselves have not advanced as much as the humans. But, I still want to look back at this thread in 10 years and say "I told you that the GTO winrate vs. 2021 500z regs would be 30bb/100!", although I fully accept that I am likely to be wrong :P

Hopefully ture GTO never playes real NL500z xD
There is one more fundamental difference between chess engines and solvers that is consequence of differences in game type. Chess engines punish every mistake you made maximally but solvers don't because of incomplete information thing in poker. In that sense engines are much stronger then solvers.

I think next big step in slover/bot development will be programming bots that can gather information and exploit with that information. This is probably quite a bit harder task and seem to me that academic community lost interest in poker because computers already reached superhuman level, so we might wait for sometime for that.

I just played $1/3$ live poker with one straddle up to 5x available from any position. $400 dollar max buy in (133bb). The game was so sloppy and messy. It was poker chaos at its best. Numerous different stack sizes. Frequently the ugh would straddle for 2x and frequently folks would straddle the button pot sized, 11$. When a straddle was on the game seemed to get looser and more passive so we’d see 5 or 6 way limped pots or if not we’d see raised and multi-way pots with drastically different raise sizes. Some people min raised. I kept increasing my raise size until I got folds, which I found was 7x plus the limps.

Poker is so complex. It’s hard to believe we can program a computer to control for all possible variables.

I had a primitive pio-like solver in 2008 already as did other people probably. (screenshot: https://jeskola.net/sv2.png). It was very slow though - that one probably took an hour or two.

I think you mean 2015 because the first public BETA of Jesolver was in december 2015 .