1.g4 opening is losing?

[Ovyron]

If it requires a "tablebase-like" draw, it's going to be very hard to find.

That's not how it works (drawn positions have plenty of drawing lines, winning positions have plenty of winning lines, you don't need to find a single line that does something like in cursed wins.)

[jp]

We don't know whether drawing would require very precise moves. Maybe. Maybe not. You can't just spout some vague generalization about "how it works" for a specific position.

There must be many endgame examples where there's no margin for error for the defending side.

[zullil]

You still think it's drawn zullil? What do you think someone could do to find a single drawing line for white? (to at least put the ball on the other camp, people having to find a black win against that line)

Yes, I still believe 1. g4 is a theoretical draw. This is simply a stubbornly held personal belief.

As for how to find a drawing line, wouldn't this be a good "machine learning" project? That is, have a neural net play White starting from 1. g4, and let it discover, i.e., learn, how to survive as long as possible. Not sure I see any better approach ...

[SheikhYerbouti]

You still think it's drawn zullil? What do you think someone could do to find a single drawing line for white? (to at least put the ball on the other camp, people having to find a black win against that line)

Yes, I still believe 1. g4 is a theoretical draw. This is simply a stubbornly held personal belief.

As for how to find a drawing line, wouldn't this be a good "machine learning" project? That is, have a neural net play White starting from 1. g4, and let it discover, i.e., learn, how to survive as long as possible. Not sure I see any better approach ...

It could well be a draw. Best to apply machine learning to find out. Would also be interesting to see how a program trained from g4 only would generalize well across openings and throughout the game.

Another idea, SF plays itself but with extreme contempt value for white.

[Ovyron]

Another idea, SF plays itself but with extreme contempt value for white.

It's trivial to modify Stockfish to make it think that a draw is as valuable as mating black, I don't see contempt getting more extreme than that. And it doesn't help finding anything.

It doesn't prove anything, but I've only seen people that thought 1.g4 could be saved jumping towards this side of the fence (me included, and believe me, I'm way more stubborn than zullil), but I've never seen anyone that thought 1.g4 was lost to jump to the other side. This is the same path that scientific discoveries follow, first they're ignored, then ridiculed, then fought, and finally accepted.

[jp]

The obvious next step is to get Hai to play White against someone.

[Uri Blass]

Another idea, SF plays itself but with extreme contempt value for white.

It's trivial to modify Stockfish to make it think that a draw is as valuable as mating black, I don't see contempt getting more extreme than that. And it doesn't help finding anything.

It is not trivial.
The problem is that draw or not draw is dependent on the history of the game.

With one history of the game you get a draw by repetition.
With another history of the game you get a win for black.

Now you may learn wrongly some position is a draw and choose a losing line only because with a different history the same position is a draw.
We never can trust draw scores by most chess engines because of this type of problem.

[Zenmastur]

You still think it's drawn zullil? What do you think someone could do to find a single drawing line for white? (to at least put the ball on the other camp, people having to find a black win against that line)

Yes, I still believe 1. g4 is a theoretical draw. This is simply a stubbornly held personal belief.

As for how to find a drawing line, wouldn't this be a good "machine learning" project? That is, have a neural net play White starting from 1. g4, and let it discover, i.e., learn, how to survive as long as possible. Not sure I see any better approach ...

And if you are wrong... someone has wasted a life time trying to find a draw that isn't there. I think the time would be better spent trying to find a better method than using neural nets.

From what I've seen they couldn't find their ass with both hands.

E.g. use a net to find a mate in the problem I posed for Ovyron. After if finds a mate you can send a note to my great-great grandchildren!

[Ovyron]

Now you may learn wrongly some position is a draw and choose a losing line only because with a different history the same position is a draw.
We never can trust draw scores by most chess engines because of this type of problem.

Well, what can I say? In 2014 I lost a certain game against a guy named Antares because of this, I used Stockfish with learning to solve the position to 0.00, and claimed I could draw it against god (while he claimed some crazy mate in 73), and then we reached the position with a different move history and I was lost and had to resign, and we've spent the last 6 years talking about that game (whether he defeated me, or he defeated this bug, that I could have drawn it without it, or lost anyway...)

So I'd rectify and say it's not trivial, but still, building an engine that thinks a draw is +INFINITY instead of 0.00 doesn't seem to make a difference for 1.g4.

[zullil]

You still think it's drawn zullil? What do you think someone could do to find a single drawing line for white? (to at least put the ball on the other camp, people having to find a black win against that line)

Yes, I still believe 1. g4 is a theoretical draw. This is simply a stubbornly held personal belief.

As for how to find a drawing line, wouldn't this be a good "machine learning" project? That is, have a neural net play White starting from 1. g4, and let it discover, i.e., learn, how to survive as long as possible. Not sure I see any better approach ...

And if you are wrong... someone has wasted a life time trying to find a draw that isn't there. I think the time would be better spent trying to find a better method than using neural nets.

From what I've seen they couldn't find their ass with both hands.

E.g. use a net to find a mate in the problem I posed for Ovyron. After if finds a mate you can send a note to my great-great grandchildren!

I was careless with my wording. I shouldn't have written "find a drawing line." I don't think any of us are going to have a proof about 1. g4 in this lifetime. And what I intended was to train a neural net to play White from 1. g4. Not to feed the position to Lc0 and some existing net to obtain a PV.