future of top engines:how much more elo?

[Uri Blass]

A random engine will not be zero Elo.
An engine that chooses the worst possible move after a careful search might be able to hit zero.

The referenced video already explores that, and you're right, if you make Stockfish sort the moves from best to worse, and make the worst one every time, it'll achieve the lowest possible elo. It'll be interesting if someone can come up with a weaker playing algorithm.

If you give this weakest opponent 0 elo, then Random has 269.8 elo.

This is clearly wrong.
It is easy to lose every game against the opponent that make the worst move in every move or against the random player.
You simply win materail against it and get a winning position and later use the material advantage that you have to force the opponent to checkmate.

Note that I do not consider in this post another way that is more simple and more convincing to lose(resigning in the first move of the game).

[Laskos]

A random engine will not be zero Elo.
An engine that chooses the worst possible move after a careful search might be able to hit zero.

The referenced video already explores that, and you're right, if you make Stockfish sort the moves from best to worse, and make the worst one every time, it'll achieve the lowest possible elo. It'll be interesting if someone can come up with a weaker playing algorithm.

If you give this weakest opponent 0 elo, then Random has 269.8 elo.

This is clearly wrong.
It is easy to lose every game against the opponent that make the worst move in every move or against the random player.
You simply win materail against it and get a winning position and later use the material advantage that you have to force the opponent to checkmate.

Note that I do not consider in this post another way that is more simple and more convincing to lose(resigning in the first move of the game).

I think even "Andworst" - the worst Andscacs --- is weaker than Worstfish, IIRC it is 400+ Elo points below Random. I think some clever schemes of losing to the Random player can make an engine being 1000+ Elo points weaker than Random. Well, aside this instantly resigning scheme (we are not talking of these here).

[MikeB]

A random engine will not be zero Elo.
An engine that chooses the worst possible move after a careful search might be able to hit zero.

The referenced video already explores that, and you're right, if you make Stockfish sort the moves from best to worse, and make the worst one every time, it'll achieve the lowest possible elo. It'll be interesting if someone can come up with a weaker playing algorithm.

If you give this weakest opponent 0 elo, then Random has 269.8 elo.

This is clearly wrong.
It is easy to lose every game against the opponent that make the worst move in every move or against the random player.
You simply win materail against it and get a winning position and later use the material advantage that you have to force the opponent to checkmate.

Note that I do not consider in this post another way that is more simple and more convincing to lose(resigning in the first move of the game).

I think even "Andworst" - the worst Andscacs --- is weaker than Worstfish, IIRC it is 400+ Elo points below Random. I think some clever schemes of losing to the Random player can make an engine being 1000+ Elo points weaker than Random. Well, aside this instantly resigning scheme (we are not talking of these here).

it's an interesting human dynamic at work here as the thread "future of top engines:how much more elo?" has morphed into "how low can we go?"

[todd]

A fun engine idea: Start out playing like full-strength Stockfish (in order to win some material) and then use near stalemate positions and self-smothering to force the opponent to checkmate it.

[Leo]

It would be interesting to know how much a 75-100 ply search and beyond would increase elo.

[mwyoung]

after the latest stf6 komodo 8 etc, i wonder how much elo more can be achieve via programming,on the same hardware of course.
100,200,300 elo more? it could be a poll but anyway just a "food of thought"
my personal feeling not more than 100 elo. more? it would be a miracle
dont forget pls on the same hardware! lets say a medium pc 2 core etc..

I been seeing the same predictions for almost 20 years. And progress has not stopped yet given equal hardware. I love pulling these predictions, and playing the programs from 5 years ago, 10 years ago, 15 years ago.... And watching the modern programs thrash the older programs by hundreds of elo points.

These prediction errors are made because people generally do not understand the vastness of the game tree in chess. And think chess is solvable.

One of the funniest things I have seen is someone claiming if they could use a 100,000,000 CPU and GPUs chess could be solved.

In 5 years I will be doing the same with todays programs...

[Leo]

after the latest stf6 komodo 8 etc, i wonder how much elo more can be achieve via programming,on the same hardware of course.
100,200,300 elo more? it could be a poll but anyway just a "food of thought"
my personal feeling not more than 100 elo. more? it would be a miracle
dont forget pls on the same hardware! lets say a medium pc 2 core etc..

I been seeing the same predictions for almost 20 years. And progress has not stopped yet given equal hardware. I love pulling these predictions, and playing the programs from 5 years ago, 10 years ago, 15 years ago.... And watching the modern programs thrash the older programs by hundreds of elo points.

These prediction errors are made because people generally do not understand the vastness of the game tree in chess. And think chess is solvable.

One of the funniest things I have seen is someone claiming if they could use a 100,000,000 CPU and GPUs chess could be solved.

In 5 years I will be doing the same with todays programs...

I agree.

[Uri Blass]

after the latest stf6 komodo 8 etc, i wonder how much elo more can be achieve via programming,on the same hardware of course.
100,200,300 elo more? it could be a poll but anyway just a "food of thought"
my personal feeling not more than 100 elo. more? it would be a miracle
dont forget pls on the same hardware! lets say a medium pc 2 core etc..

I been seeing the same predictions for almost 20 years. And progress has not stopped yet given equal hardware. I love pulling these predictions, and playing the programs from 5 years ago, 10 years ago, 15 years ago.... And watching the modern programs thrash the older programs by hundreds of elo points.

These prediction errors are made because people generally do not understand the vastness of the game tree in chess. And think chess is solvable.

One of the funniest things I have seen is someone claiming if they could use a 100,000,000 CPU and GPUs chess could be solved.

In 5 years I will be doing the same with todays programs...

In theory if people write the best possible software for some hardware then there cannot be improvement only by software even without solving chess.

The improvement only by software show something not about chess but about programming and it shows that the chess software of the best engines is not close to being perfect not in the meaning of solving chess but in the meaning that it is impossible to get improvement only by software.

I believe that one of the reasons for big elo gap between old programs and new programs is the fact that old programs are not designed to play for a draw against stronger players.

It may be interesting how much elo programs can earn by knowing the elo of the opponent against the default version.

For example suppose that you know that your opponent is 200 elo stronger if you play normally.
Can engine change their style to reduce the gap to a significantly smaller number than 200 elo based on the information(for example by changing contempt?)

Same question for the case that you know your opponent is 200 elo weaker.

It may be interesting to have some tournament with unequal time control when the participants use different contempt to see how much elo can be achieved by contempt.

I believe programmers may do better than contempt but unfortunately I do not see tournaments to test it.

[Ovyron]

I think even "Andworst" - the worst Andscacs --- is weaker than Worstfish, IIRC it is 400+ Elo points below Random.

Ah, but the thing here is that it's not just against Random, here the ELO is calculated against the entire pool of opponents (including Stockfish 1000000 nodes, and lots of entities that play Random some of the time and Stockfish 1000000 nodes the rest of the time). Probably what we'd need is an entity that plays for a bit to check what kind of opponent it's playing, then ensure it gets checkmated either by playing lousy or by playing great and forcing opponent to checkmate it, but naive solutions wouldn't work.

I don't like the idea of negative elo, so Dann's idea of calibrating the weakest chess entity to elo 0 would help us in these discussions as we could talk about some elo according to this scale that isn't arbitrary, so 4500 elo where weakest entity is 0 elo is meaningful. And, no, resigning every game before game starts is bypassing the problem.

The project is still ingoing, Tom 7 invited people to suggest their own playing algorithms on the video comments, but this is just all for fun, as Stockfish represents the Alphabeta algorithm, there's many other ways to pick chess moves, as long as you don't care about actually winning the game.

[Uri Blass]

I think even "Andworst" - the worst Andscacs --- is weaker than Worstfish, IIRC it is 400+ Elo points below Random.

Ah, but the thing here is that it's not just against Random, here the ELO is calculated against the entire pool of opponents (including Stockfish 1000000 nodes, and lots of entities that play Random some of the time and Stockfish 1000000 nodes the rest of the time). Probably what we'd need is an entity that plays for a bit to check what kind of opponent it's playing, then ensure it gets checkmated either by playing lousy or by playing great and forcing opponent to checkmate it, but naive solutions wouldn't work.

I don't like the idea of negative elo, so Dann's idea of calibrating the weakest chess entity to elo 0 would help us in these discussions as we could talk about some elo according to this scale that isn't arbitrary, so 4500 elo where weakest entity is 0 elo is meaningful. And, no, resigning every game before game starts is bypassing the problem.

The project is still ingoing, Tom 7 invited people to suggest their own playing algorithms on the video comments, but this is just all for fun, as Stockfish represents the Alphabeta algorithm, there's many other ways to pick chess moves, as long as you don't care about actually winning the game.

You cannot know what type of opponent you play based on the first moves because the opponent may also play strong in the opening only in order to lose later(same strategy that is best to lose against the random player).

Rating is dependent on the pool of players and basically you can score better against A only if you are ready to score worse against B