Fritz 18: Evolving Genius

[Milton]

I bought the program last night and so far I am enjoying the evolving genius mode. I like that it gives indirect hints and announces in some positions that a decisive move exists for humans. This may render the feature too easy for some folks, but I need all the help I can get.
I think I found one minor bug in the GUI. On my computer the "Chess Benchmark" feature does not work at all. By comparison, in the Fat Fritz 2 GUI, which I also have, the benchmarking feature does work. Can anyone else with Fritz 18 confirm whether the Chess Benchmark feature works for them? Thanks.

[AdminX]

Okay, so I just downloaded and installed Fritz 18. Then I played a quick test game to test out some of the new features. I did find it helpful as a sort of coach, however it also felt like I was cheating and being helped by some nearby kibitzer shouting out advice I did not ask for. Of course the game was installed with the default setting which I had not yet changed. I think the new interface and AI will pair nicely with the Dragon by Komodo engine for players that wish to prepare for human games.

Were you able to play around with the Evolving Genius thing yet? How does it work in practice, and does it work for other engines like Dragon?

Okay, so I changed engine to Dragon, then went to 'Easy Game'. I could not see it load any engine, so I am not sure if it even used Dragon. I then used Process Explorer to see if Dragon was loaded and did not see it.



UPDATE: After the 2nd game completed, I clicked on a random move number from the game and then clicked on infinite analysis. Sure enough it then loaded Dragon which I could see in Process Explorer. This tell me that is not using Dragon in that mode, maybe just some build in version of Fritz as when I have Fritz 18 loaded as the default engine I can see it loaded when going into easy game.

[Madeleine Birchfield]

Okay, so I changed engine to Dragon, then went to 'Easy Game'. I could not see it load any engine, so I am not sure if it even used Dragon. I then used Process Explorer to see if Dragon was loaded and did not see it.



UPDATE: After the 2nd game completed, I clicked on a random move number from the game and then clicked on infinite analysis. Sure enough it then loaded Dragon which I could see in Process Explorer. This tell me that is not using Dragon in that mode, maybe just some build in version of Fritz as when I have Fritz 18 loaded as the default engine I can see it loaded when going into easy game.

So it would seem that, unlike what Cornfed was saying earlier in the thread, Evolving Genius is a feature that can only be used with the Fritz 18/Gingko engine. Disappointing, to say the least.

So long as the Fritz 18 engine still uses a handcrafted evaluation, any Fritz 18-exclusive features are not worth it in my opinion. You could find better engines with better evaluations, especially in the opening and middlegame, as they use neural networks in their evaluations. For a sparring partner or analysis tool, an engine using neural networks would end up playing more human-like moves and suggest more intuitive lines than an engine using only a handcrafted evaluation, and thus would be far more valuable to the average human player than the Evolving Genius feature.

[reflectionofpower]

I haven't even gotten around to playing around with it in full.

[Cornfed]

Okay, so I changed engine to Dragon, then went to 'Easy Game'. I could not see it load any engine, so I am not sure if it even used Dragon. I then used Process Explorer to see if Dragon was loaded and did not see it.



UPDATE: After the 2nd game completed, I clicked on a random move number from the game and then clicked on infinite analysis. Sure enough it then loaded Dragon which I could see in Process Explorer. This tell me that is not using Dragon in that mode, maybe just some build in version of Fritz as when I have Fritz 18 loaded as the default engine I can see it loaded when going into easy game.

So it would seem that, unlike what Cornfed was saying earlier in the thread, Evolving Genius is a feature that can only be used with the Fritz 18/Gingko engine. Disappointing, to say the least.

So long as the Fritz 18 engine still uses a handcrafted evaluation, any Fritz 18-exclusive features are not worth it in my opinion. You could find better engines with better evaluations, especially in the opening and middlegame, as they use neural networks in their evaluations. For a sparring partner or analysis tool, an engine using neural networks would end up playing more human-like moves and suggest more intuitive lines than an engine using only a handcrafted evaluation, and thus would be far more valuable to the average human player than the Evolving Genius feature.

Actually...

If you look back, you asked and I said " I would not know...it hasn't been released yet. I would presume it is a function of the GUI".

In any case, The feature is for Humans to PLAY or 'SPAR' AGAINST and for the engine to play strongly, but not perfectly and to give you a chance in some games to spot opportunities an engine normally would not give you the chance to....kind of like a local IM and former roommate of mine who lives down the road could end up doing....or indeed did in our OTB game last weekend. Humans, even very good ones...miss things, but you have to spot them and convert.

I am not sure NNUE evals vs human evals really matter for this purpose. A 3100 elo hand crafted eval engine (instead of a 3500 NNUE engine) perhaps mimicking a strong human who lets his guard down a little once in a while...that's good enough of a sparing partner for 99% of human players, don't you think?

Now, I haven't tried it yet...but I have played some of those old computers and engines (think Sargon or Chess Challenger for example) and I think the experience (if you turn off the annoying TALK feature) will likely quite a bit better. Again, not having tried it yet, I have to add "I would not know"...

[Odd Gunnar Malin]

Wonder if there is some misunderstanding here. I think this new feature is a learning tool and you use it like that. Probably all hints/moves etc. comes from an internal engine. And I don't think it really matter which, because the logic is a GUI thing. Someone who disassembled Centaur said that it didn’t matter if you replaced Stockfish with another engine, the playing style would be the same. The selection of moves was from outside the engine.

For your training games you do as before, e.g. go into ‘Classic menu’ and play rated games against whatever engine you have installed.

[Madeleine Birchfield]

I am not sure NNUE evals vs human evals really matter for this purpose. A 3100 elo hand crafted eval engine (instead of a 3500 NNUE engine) perhaps mimicking a strong human who lets his guard down a little once in a while...that's good enough of a sparing partner for 99% of human players, don't you think?

The issue is that engines with handcrafted evaluations simply do not mimic a strong human, regardless of what elo it is at. They play like superhuman calculators who are masters at tactics, but absolutely suck at positional play and are crap in the opening. No human plays like that. Chess players ranging from your average club player to GMs have complained for over a decade and a half how engines using handcrafted evaluation simply do not play like a human.

It is the engines with neural networks that mimic a strong human who lets their guard down a little once in a while, and the larger the neural network, the better the approximation. So NNUE evals are better than handcrafted evaluations, and deep CNN evals like Leela and ClassicAra are better than NNUE engines. It's the same reason why Maia Chess was so widely praised by the chess community when it came out last year, because it uses neural networks and were able to mimic human performance (even though it mimics lower-rated players rather).

[Odd Gunnar Malin]

I think I found one minor bug in the GUI. On my computer the "Chess Benchmark" feature does not work at all. By comparison, in the Fat Fritz 2 GUI, which I also have, the benchmarking feature does work. Can anyone else with Fritz 18 confirm whether the Chess Benchmark feature works for them? Thanks.

This feature doesn't work here either.
You have a program called 'Fritz Chess Benchmark.exe' in the program folder, was it just starting this tool. I uninstalled Fritz 17 before I installed Fritz 18 so I can't check. By the way, is it allowed to give away the old (downloaded) program, I didn't buy an upgrade.

[Milton]

I think I found one minor bug in the GUI. On my computer the "Chess Benchmark" feature does not work at all. By comparison, in the Fat Fritz 2 GUI, which I also have, the benchmarking feature does work. Can anyone else with Fritz 18 confirm whether the Chess Benchmark feature works for them? Thanks.

This feature doesn't work here either.
You have a program called 'Fritz Chess Benchmark.exe' in the program folder, was it just starting this tool. I uninstalled Fritz 17 before I installed Fritz 18 so I can't check. By the way, is it allowed to give away the old (downloaded) program, I didn't buy an upgrade.

Yes, "Fritz Chess Benchmark.exe" exists in both folders "Chessprogram17" and "Chessprogram18" on my computer. Directly clicking on either of those files causes the benchmark to run. But clicking on the "chess benchmark" icon in the main GUI only works for the Fat Fritz 2, but not for the Fritz 18 GUI. Thanks for your response. I have sent an email to Chessbase to let them know about this glitch.

I honestly do not know the answer to your questions about whether it is allowed to give away your old program.

[BrendanJNorman]

The issue is that engines with handcrafted evaluations simply do not mimic a strong human, regardless of what elo it is at. They play like superhuman calculators who are masters at tactics, but absolutely suck at positional play and are crap in the opening.

I disagree with this, but feel that you are getting your tools mixed up. Let me explain...

1. There is a large difference between an engine that is *aimed* at being as strong as possible (Elo chasing) and an engine *aimed* at simulating a human player.

Of course, a hand-crafted eval aimed at being strong as possible will appear primitive compared to an NN eval with the same aim.

But if you are deliberately aiming to simulate a human opponent (create a sparring partner), there is no better path (at this point) than HCE.

This is because you get a whole smorgasbord of values you can tinker with (piece values, king safety, pawn structure, material value and so on), plus depth/NPS settings, plus randomization, plus so much more, while an NN eval is simply a black box where you have no idea *why* it is choosing certain moves.

2. The "superhuman calculation/tactician" can be dumbed down easily with a depth/NPS limit setting or similar UCIElo type settings.

3. HCE *only* "sucks at positional play" compared to the BEST NN engines, but is still superior to the BEST human players - and since we are trying to replicate human play, this is a non-issue.

4. "Crap in the opening" easily solved with polyglot book made up of the games of sub-2500 players and limited to about 10 moves.

Point being, a HCE engine with highly biased eval (which is one big human weakness) such as lower/higher default material value, over/under estimation of pawn structure, weaker than usual king safety, etc - whatever style you're aiming for - with a custom opening book and limited search (by whatever means you choose), will play much more human-like than an NN engine.

There is great work being done to produce weakened nets, but they aren't there yet. The positional play is still far too strong compared to even master-level humans.

And I say this as someone who plays blitz against master-level humans daily.