Chess System Tal Extreme, draw rate very low

[cpeters]

I understand that most programmers just do engine programming for sake of internal competition and to be an esteemed part of the developer community - but isn't it still a valid question what users would be interested to see as new developments?

Yes/maybe. But you might want to communicate or verbalise these wishes through proper channels and to the right recipients. Think of Speaker's Corner - how far ramblings will travel or can be heard?

I'm using Linux since twenty years and in the case of free and open source software I learnt that you as a user have no rights at all, besides what the license is granting you - and that's already generous.

You may file: bug-reports, feature-requests and so on. You may help with: documentation, translation, provide code (to fix bugs/enhance usability) or else.

That's all. Otherwise you've to shut up or hand over money (to get your will so to speak).

[towforce]

We're being told in this thread that CSTE lowers the draw rate by making speculative sacrifices.

This is good, and also contains an important "life lesson" about how you can behave if you're stuck in a situation you don't like.

An even better way to lower the draw rate would be to be to prefer complex positions. This would:

1. Confuse both humans and ANNs, who would no longer be able to understand positions

2. Throttle search. Chess engines' search depth drops dramatically in complex positions where there are a lot of attacks possible in the near future

The downside is that the games would be less fun to watch or play in than an engine producing swashbuckling sacrifices.

[Aleks Peshkov]

2. Throttle search. Chess engines' search depth drops dramatically in complex positions where there are a lot of attacks possible in the near future

I'd say opposite. The more forced lines the easier for alpha-beta to prune futile lines.

My engine has similar rating and have seen double kill from CSTE. The problem I suspect: trained on common positions my NNUE evaluation does not understand Tal Extreme game positions. It looks like my engine values unbalanced positions poorly and thus make poor move choices when game is not one side decided. Training more in unexplored territory should cure from temporary blindness.

[towforce]

2. Throttle search. Chess engines' search depth drops dramatically in complex positions where there are a lot of attacks possible in the near future

I'd say opposite. The more forced lines the easier for alpha-beta to prune futile lines.

That's the trouble: in complex positions you won't find forced lines: what you will find is that for the next several moves, there are a huge number of attacking options, none of which lead to quiescence: the engine has no option but to search through this plethora of combinations in full detail.

In such positions, I have personally seen declared search depth fall to something like 25% of its normal level for that engine.

Killing both an engine's ability to understand a position and it's ability to search would greatly improve the chances of it making a fatal error.