Hale-Bopp wrote on 12/29/20 at 10:23:09:

He seems to be saying about the positions after 16. h3 that the neural network engines are less pessimistic than traditional engines (White is not losing or clearly worse as with Stockfish 12), and that anyway a human would rather play White because figuring out how to make progress (notwithstanding the engine evaluation) as Black is really difficult and Black has chances to go wrong.   He posted a blitz game with him against GM Tang where Tang couldn't find the right moves over the board.  To prepare to extract the theoretical Black computer advantage, he implicitly argues, you would need to prepare (e.g., play a few training games on both sides with a friend or computer to get the hang of how to make progress as Black) and who is going to do that for a line they will never see?  

For me, relying on his argument is not enough to go back to trying the Moeller as White, even for blitz.  Why bother?  White would have to do some work to understand the nuances of this position too and there are easier fish to fry.  In the US, even Class A players are totally booked up on the Moeller unlike some obscure but less sound line in say the Belgrade Gambit or the Vienna Game.  For an e4 player looking to pose problems after e5, why not put in your work elsewhere?

As Black, this may make me favor 9 ... Ne5 (and the 11 ... f5 line) over 9 ... Bf6 . . .  as there is less homework involved (and it is easy enough to not fall for the Ne5 Moeller traps published 75 years ago in an Irving Chernev book).

As a larger proposition, using the NN engines to reevaluate lines like this would make sense if there was a way to provide evaluations of human playability at different levels (holding the game stats to the side because those only come after the line has been played and studied).  Obviously, titled players are titled because they can understand this but I wonder if there is a role for engines here too (maybe there is and I don't know about it).

an ordinary chessplayer wrote on 12/29/20 at 17:29:49:

A couple of historical bits:  Fine gave 11...f5 as "!" and as leading to a slight advantage for Black.  50 years later, Nunn gave it as leading to a slight advantage for White (hmm).

I don't know the article, but in a recent YouTube video Braun recommends

8. ...Bxc3 9.d5 Bf6 10.Re1 Ne7 11.Rxe4 d6 12.Bg5 Bxg5 13.Nxg5 h6 14.Qe2 hxg5 15.Re1 Be6 16.h3!?, and then for example 16. ...c6 17.dxe6 f6 18.Bd3.

Honestly, that doesn't seem too exciting to me. Also, when you check it with an engine, Black seems to be absolutely fine (clearly better, actually). The position might be a bit tricky, but it's nothing too dangerous for Black and objectively it seems to be nothing at all for White.

It would also be kind of interesting if Braun has anything to show against the many other good options for Black:

11. ...0-0 12.d6 cxd6 seems to be absolutely fine for Black;

even 13. ...0-0 14.Nxh7 is still a draw, even if this is really dangerous if Black doesn't know it.

9. ...Ne5 is of course also more than fine,
10.bxc3 Nxc4 11.Qd4 O-O (11... f5 seems also good) 12. Qxe4 Nd6 13.Qf4 Re8 14.Ba3 Re4 seems pretty forced and doesn't look too attractive for White, either.

Nowadays, anyone can check this out quickly. Does Braun have anything halfway convincing to show here? 

In CBM 199 there is an article by Christian Braun (perhaps the German IM) about a new idea in the ancient Møller Attack (1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 4. c3 Nf6 5. d4 ed 6. cd Bb4+ 7. Nc3 Nxe4 8. 0-0 Bxc3 9. d5).  I wonder if someone could give a basic indication of what that's about.