Dear Google AlphaZero chess team.

[Milos]

Thank you for enlightening me, I had absolutely no idea whatsoever that Pablo could draw against 3300 ELO engines so easily.

A0 is well below 3300Elo at 5min/game TC as shown in the very same paper about it.
Again you do demonstrate you don't have much clue about A0, beside going through 10 cherry-picked games.

[MonteCarlo]

Well, re: the "as shown in the very same paper" bit, let's not go too crazy

Nothing in the paper indicates that A0 loses several hundred points relative to SF at game in 5 minutes (for that matter, such a time control is not discussed at all).

The only part of the paper remotely close to talking about that is the part that discusses scaling with time, and the results do indeed show that at very fast time controls, A0 is weaker than SF, but their strength curves meet before 10^0 seconds per move, and A0 is already stronger than SF at 10^0 seconds per move, which is still a faster game than game in 5 minutes for the vast majority of games.

Now, it's somewhat reasonable to suppose (as has been stated ad nauseum in the seemingly hundreds of threads about this) that SF would get some boost from getting to use normal time management in a time control like game in 5 minutes relative to using a fixed time per move, but the extent to which that would help it against A0 is a somewhat speculative matter, and certainly there's nothing about that in the paper

In short, is it possible that A0 is several hundred points weaker than SF at G/5? Sure, it's conceivable, I suppose, but claiming that you could somehow read or deduce that from the paper is a bit much

[FWCC]

Much respect Pablo,God Bless
We know what you can do.








FWCC

TAL WAS CORRECT

[Milos]

Well, re: the "as shown in the very same paper" bit, let's not go too crazy

Nothing in the paper indicates that A0 loses several hundred points relative to SF at game in 5 minutes (for that matter, such a time control is not discussed at all).

The only part of the paper remotely close to talking about that is the part that discusses scaling with time, and the results do indeed show that at very fast time controls, A0 is weaker than SF, but their strength curves meet before 10^0 seconds per move, and A0 is already stronger than SF at 10^0 seconds per move, which is still a faster game than game in 5 minutes for the vast majority of games.

Now, it's somewhat reasonable to suppose (as has been stated ad nauseum in the seemingly hundreds of threads about this) that SF would get some boost from getting to use normal time management in a time control like game in 5 minutes relative to using a fixed time per move, but the extent to which that would help it against A0 is a somewhat speculative matter, and certainly there's nothing about that in the paper

In short, is it possible that A0 is several hundred points weaker than SF at G/5? Sure, it's conceivable, I suppose, but claiming that you could somehow read or deduce that from the paper is a bit much

Several hundred Elo is not devised from the paper and I never claimed that. I claimed (and that is totally clear from the paper) that A0 is much weaker than 3300Elo (assuming that it is A0 strength at 1min/move TC) at 5min/game TC.
Several hundred Elo weaker than SF is due to the one important fact that you forgot in your comment. At 1min/move in the paper SF has been severely handicapped to up to probably 100Elo.
However, already at 0.3sec/move even that severely handicapped SF becomes stronger than A0.
With 5min/game TC, those handicap effects on SF become much, much more pronounced and they can easily go from 100Elo to 200Elo or more.

[Ovyron]

A0 is well below 3300Elo at 5min/game TC as shown in the very same paper about it.

Of course, I can defeat A0, the one that defeated Stockfish 8, the whole week, and twice on Fridays, on a 5min/game TC.

The game would go like this:

I play:

1.e4 (best by test, A0 hasn't learned well!)

A0 replies:

1...e5?? (blunder, everyone knows the Sicilian is best)

I play:

2.Nf3

A0 replies:

2...Nc6?? (blunder, the Petroff is more solid)

I play:

3.Bb5

A0 replies:

3...Nf6!? (innacuracy, 3...a6 is better first)

I play:

4.d3

A0 replies:

4...Bc5!? (I'll just stop commenting on this nonsense...)

I play:

5.Bxc6

A0 replies:

5...dxc6!?

I play:

6.O-O!!

A0 replies:

Alpha Zero forfeits on time, white is victorious!

Haha! See? A0 plays a 1 min/move, I ask for game in 5 minutes, it doesn't have time to even make its 6th move...

That's what you meant by Pablo having decent chances, right?

Otherwise, there's no such thing as "5 min/game" when A0 plays at fixed time per move.

There's no such thing as locking the position and beating A0 on time when A0 always makes a move in the specified time.

There's no such thing as decent chances when Pablo is only able to win by locking the position and winning on time.

And there's no such thing as DeepMind implementing time management just so that Pablo has a chance to lock the position and beat A0 on the clock.

[Milos]

A0 is well below 3300Elo at 5min/game TC as shown in the very same paper about it.

Of course, I can defeat A0, the one that defeated Stockfish 8, the whole week, and twice on Fridays, on a 5min/game TC.

Haha! See? A0 plays a 1 min/move, I ask for game in 5 minutes, it doesn't have time to even make its 6th move...

That's what you meant by Pablo having decent chances, right?

Otherwise, there's no such thing as "5 min/game" when A0 plays at fixed time per move.

There's no such thing as locking the position and beating A0 on time when A0 always makes a move in the specified time.

There's no such thing as decent chances when Pablo is only able to win by locking the position and winning on time.

And there's no such thing as DeepMind implementing time management just so that Pablo has a chance to lock the position and beat A0 on the clock.

You really think that Google ppl are that stupid not to know how to implement a trivial time-manager with just one category of moves for different TCs? Something that would require 10mins top to implement if you ever looked at TM code of any modern engine?
And you believe actually Google ppl didn't have time to implement it, and have chosen 1min/move TC just by chance???
Gee, I really thought you were a smarter person. What you believe seems equally convincing as believing in Santa Clause...

[Ovyron]

You really think that Google ppl are that stupid not to know how to implement a trivial time-manager with just one category of moves for different TCs?

Oh, I thought they already did? Weren't you saying that the paper said that A0 was playing at game/5 minutes? Weren't you even asserting what was A0's elo at game/5 minutes? How can this be possible if they haven't implemented a time-manager?

[Milos]

Oh, I thought they already did? Weren't you saying that the paper said that A0 was playing at game/5 minutes? Weren't you even asserting what was A0's elo at game/5 minutes? How can this be possible if they haven't implemented a time-manager?

So only what is in the paper happened, right?
How can anyone make any conclusion that is not written in the paper?
Then it seems there were only 10 different games in total that A0 won, since only these are in the paper. Since there are no other wins of A0 shown that could only mean that all other wins were identical to those 10 games shown.
You logic is simply staggering.
Normal human logic is to use something called deduction.

[MonteCarlo]

A0 is well below 3300Elo at 5min/game TC as shown in the very same paper about it.

(my emphasis added)

This claim you did make, and that cannot simply be read or deduced from the paper.

Again, I'm not disputing that it's possible that SF is stronger, even much stronger, than A0 at G/5 with Cerebellum, more cores, bigger hash, etc.

That's all quite possible, but again, none of that is "shown in the very same paper".

It's an inference you're drawing from a body of evidence that includes much, much more than the paper.

It may or may not be true; I'm not going to make any claims about how likely it is to be true, other than that it is possible.

I'm only pointing out that the claim you made in the quoted segment above was not anywhere shown in the paper. That's all

[Ovyron]

Normal human logic is to use something called deduction.

Wow! Deduction! That's a completely new concept to me. So let's see if I can do something useful with my new deductive powers...

Suppose that DeepMind created another Neural Network that was responsible to manage time in time control games, it would play random games against itself, and soon learn that too much time used would lose on time.

Then it would learn that playing too fast dramatically decreases the strength of its moves (say, A0 with 2:30 minutes remaining on the clock would play half as strong as one with a few seconds at the end of a 5 minute game.)

Eventually you'd get the best time managers and A0 would be equipped with the best one they were able to find, for her match against Pablo.

In the actual match, the whole time management wouldn't even make any difference most of the time, since A0 wouldn't allow Pablo to lock the position in the first place...

But eventually, and with determination (and so far no draws or wins, just loses for Pablo) Pablo gets his holy grail, and locks the position! Then he moves and premoves as fast as he can, trying to flag Alpha Zero, until...

Coming to this point, is it your claim that Pablo would be able to move faster than the best time management DeepMind could build, and by "decent chances" do you mean a single win after hundreds of lost games?