Why does TCEC use unbalanced opening?

[Marcus9]

I understand that the draws at this level are very high, But I don't find unbalanced openings interesting.
I'd rather see two superfinals, Where in one an engine has to play with time odd, for example 120'+10" vs 12'+1" or 90'+30" vs 9'+3".
This would allow us to have much more information:
- How is good an engine to scaling
- Fewer draw with good quality of chess (even in opening)
- We can see how win the same opening with reverse color

Also, the time to play the 2 superfinals will only be 10% longer than playing 1 at equal time

It may sound strange but it is no more strange to try unbalanced openings

[Marcus9]

Maybe also do a third superfinal with both sides with equal time but blitz (12'+1" vs 12'+1" or 9'+3" vs 9'+3"), For more information on scaling

[Jouni]

I think it's almost impossible to win SF anymore from balanced opening with TCEC harware! Chess is practically solved now.

[Marcus9]

I think it's almost impossible to win SF anymore from balanced opening with TCEC harware! Chess is practically solved now.

Not with time odd! Otherwise it would make no sense to use lomg TC

[Cornfed]

If an engine cannot take an 'edge' or unbalanced position and turn it into a winning game...it does not deserve to be considered a good engine. Simple.

[glennsamuel32]

https://tcec-chess.com/articles/TCEC_Openings_FAQ.html

[Alayan]

If an engine cannot take an 'edge' or unbalanced position and turn it into a winning game...it does not deserve to be considered a good engine. Simple.

You got it wrong. You can only win a game of chess from a balanced position if your opponent make mistakes. You can play more challenging moves to increase the likelihood of mistakes by the opponent, you can be better at spotting and exploiting the mistakes, but if your opponent is strong enough, it's not going to be enough.

The next level is being able to emulate your opponent and using adversarial search techniques to find which moves are most likely to lead to a win. This is very narrow, as it needs significant code and a big computing edge (it works best when you can test a lot of variations deeper than the engine you're playing against is seeing). This method accepts playing moves that will score worse against a large selection of opponents in favor of maximizing the score against one opponent.

Current engines most likely can be defeated quite reliably from the start position with both colors using adversarial search techniques, at TCEC conditions.

Then, there are positions that an engine with given hardware, TC, settings, will never lose (or fail to convert as a win) against any possible sequence of moves. We are not quite there for the start position but each progress in elo or hardware upgrade increases the number of such "practically solved" positions.

Now, step back and consider that we are not pitting an engine against a 2030 super-engine, but two 2020 engines of relatively similar strength. Balanced openings will be extremely drawish and have little use to tell you which engine is stronger. It also makes for a snorefest if you watch the games.

[Marcus9]

https://tcec-chess.com/articles/TCEC_Openings_FAQ.html

I find the match with time odd as a better solution.

Quote:
"you will see some openings where one color has no realistic winning chances and the game is all about successfully holding a draw from an almost desperate position."

If 10:1 time odd it'not enought TCEC can use also 30:1, I think it is more interesting to approach a difference of 200 elo, to have odds of winning similar to draws.

Under these conditions an engine can show how to win an opening in reversed colors and show how to hold on in the second superfinal

[Graham Banks]

I'd like to see TCEC return to using balanced openings, with perhaps a 6 moves per side limit.

[Laskos]

I'd like to see TCEC return to using balanced openings, with perhaps a 6 moves per side limit.

And 96 draws out of 100.