# Thread: Human versus Machine

1. Originally Posted by LyudmilTsvetkov
btw., if people are not aware, these days even top human players have difficulties getting even draws
against the top engines, even with 1-2 pawns handicaps and more.
In single games or small numbers of games. We don't have very much evidence involving very top class players playing large numbers of such games, because they have better things to do with their time.

The problem with these handicap games as evidence of playing strength is the difficulty of establishing what the impact on expected score of the handicap is.

For instance if Komodo plays a match against itself, with one side playing with 2+2 and the other side playing with 10+10, what does the side with 2+2 score?

If contempt is added to the side playing with 2+2 how does this affect the result?

(I accept that contempt of the style you mention would be important at increasing the computer's chances in a game played at material odds. But that tells us nothing about games played at time odds.)

2. Originally Posted by Kevin Bonham
In single games or small numbers of games. We don't have very much evidence involving very top class players playing large numbers of such games, because they have better things to do with their time.

The problem with these handicap games as evidence of playing strength is the difficulty of establishing what the impact on expected score of the handicap is.

For instance if Komodo plays a match against itself, with one side playing with 2+2 and the other side playing with 10+10, what does the side with 2+2 score?

If contempt is added to the side playing with 2+2 how does this affect the result?

(I accept that contempt of the style you mention would be important at increasing the computer's chances in a game played at material odds. But that tells us nothing about games played at time odds.)
Dear Mr. Bonham,

unfortunately, all of your comments/claims, precisely 100%, so 6/6, are wrong.

1) you want me to play on equal terms at blitz with Komodo; who does that nowadays? Of course, this might be possible,
but you should waste an enormous amount of energy, after which there is no more fun in the game
2) if you don't know, current Komodo score in handicap matches(2 pawns, knight for pawn, pawn and tempos, etc.) versus
a pool of maybe 15 or so players with elo average about 2500-2600 is something like +30 - 0 =10(some early wins for the humans
from very favourable positions not counted in); from here to the top it is only 200 elo, so completely insignificant to change the result.
Recently, Nakamura just lost N for missing c7 pawn as black at TC 45' + increment in a rout mode.
3) just the opposite, contempt works fine under all conditions, material or time handicap alike, and that has been extensively tested by
Komodo team in their matches. Why should it be otherwise?
4) I would suggest that you just try my experiment, and then report your score. Have you recently played latest versions of Stockfish and
Komodo? Those are not the engines from 5-10 years ago. Definitely not. So I simply fail to understand why you would think 3.5/10 at that
time handicap with the engine using contempt is a bad score. Do you know of someone who scores much better under similar conditions?
Of course, a human might try to match the tops, but for that to occur TC should be at least 30'.
5) rest assured, I am an all-round player, already. I can play all kinds of positions, from extremely open to deep strategical ones,
from sharp Sicilians to QGDs and Caro-Kanns. And at a level you could hardly imagine. You can not imagine the amount of stuff,
opening, endgame, middlegame, etc., we have been jointly analysing on talkchess during the last 5 years. One certainly gains something, when
analysing a difficult position with Stockfish for a couple of hours. It is difficult for me to think which types of positions we have not been
analysing.
Still, I am sticking mostly to my preferred closed games, as they are objectively the best. The KID is, one way or another, the best possible
way to handle the game, as long pawn chains are very strong, and it is for a reason that Kasparov and Fischer almost always have chosen
KID structures. For the very same reason, top engines have difficulties with those, because they are deep ones, requiring 20-30 moves looking ahead
to see something. Top engines don't have any difficulties at all with open games, for humans this might be a problem, but actually it is trade, trade,
and the game is over, very shallow.
6) well, human elo gradually increases, so humans get something, after all, by training with the engines.
With more time, the effect will be more obvious.

Now, I have to acknowledge that I would never, ever have reached my current chess understanding, had not it not been for the existence of the top
engines. They are great pals, helping you here and there. At the same time, I have always searched for original things, new unexplored territories,
so probably not everyone gets the same out of Stockfish.

3. Originally Posted by LyudmilTsvetkov
1) you want me to play on equal terms at blitz with Komodo; who does that nowadays? Of course, this might be possible,
but you should waste an enormous amount of energy, after which there is no more fun in the game
The point is that you are making extreme claims about your playing strength but providing no way by which they can be assessed. A score against a program with a known estimated rating under controlled conditions might be one way this could be checked. Even if you could only score a few percent (which would be likely however strong you are) this could provide some evidence of playing strength, particularly if you could score in a range of openings and without relying on rehearsed lines. But all you are presenting is these handicap games which are very difficult to benchmark.

2) if you don't know, current Komodo score in handicap matches(2 pawns, knight for pawn, pawn and tempos, etc.) versus
a pool of maybe 15 or so players with elo average about 2500-2600 is something like +30 - 0 =10(some early wins for the humans
from very favourable positions not counted in); from here to the top it is only 200 elo, so completely insignificant to change the result.
Recently, Nakamura just lost N for missing c7 pawn as black at TC 45' + increment in a rout mode.
This just shows that you have no clue about ratings. +30-0=10 for the computer is 12.5% for the human. That corresponds to an ELO difference of 336 points suggesting the computer under these conditions is playing at about 2886 vs opponents of average rating 2550. Carlsen would be expected to be very close against it on this basis so your claim "from here to the top it is only 200 elo, so completely insignificant to change the result." is false.

3) just the opposite, contempt works fine under all conditions, material or time handicap alike, and that has been extensively tested by
Komodo team in their matches. Why should it be otherwise?
Feel free to provide a link to reliable evidence for this claim that does not refer to material handicaps. Because of your howler with ratings above I'll assume you don't understand anything in the absence of evidence for it.

4) I would suggest that you just try my experiment, and then report your score. Have you recently played latest versions of Stockfish and
Komodo? Those are not the engines from 5-10 years ago. Definitely not. So I simply fail to understand why you would think 3.5/10 at that
time handicap with the engine using contempt is a bad score. Do you know of someone who scores much better under similar conditions?
Of course, a human might try to match the tops, but for that to occur TC should be at least 30'.
Again you are putting things into my mouth that I didn't say and claiming I am "wrong" when you have misrepresented my comments. I did not say 3.5/10 would be a bad score. I just said it is not evidence for your claims to be over 2600 strength, especially not in the absence of evidence of benchmarking of the rating differential for the handicap and contempt arrangement.

I played a very large number of games against Stockfish 4 between 2014-5 at even time. I do not make any claims about my playing strength based on these games. I am too busy at present to start on a similar project with the latest versions but I may do so in the future. I have no interest in playing a computer at any form of odds. If someone wants to present odds games as evidence of playing strength they need to play them under controlled observed conditions and benchmark them.

5) rest assured, I am an all-round player, already. I can play all kinds of positions, from extremely open to deep strategical ones,
from sharp Sicilians to QGDs and Caro-Kanns. And at a level you could hardly imagine.
Then beat strong human players in an OTB tournament. If you are as good as you claim to be (i) it will be easy (ii) it will not take much time (iii) it will make you a lot more money than trying to flog your book online to people who can see you are a troll who makes untestable claims.

6) well, human elo gradually increases, so humans get something, after all, by training with the engines.
Irrelevant to my point that there are no examples of merely strong club players proving that they have become very strong just by playing against computers.

4. Originally Posted by Kevin Bonham
The point is that you are making extreme claims about your playing strength but providing no way by which they can be assessed. A score against a program with a known estimated rating under controlled conditions might be one way this could be checked. Even if you could only score a few percent (which would be likely however strong you are) this could provide some evidence of playing strength, particularly if you could score in a range of openings and without relying on rehearsed lines. But all you are presenting is these handicap games which are very difficult to benchmark.

This just shows that you have no clue about ratings. +30-0=10 for the computer is 12.5% for the human. That corresponds to an ELO difference of 336 points suggesting the computer under these conditions is playing at about 2886 vs opponents of average rating 2550. Carlsen would be expected to be very close against it on this basis so your claim "from here to the top it is only 200 elo, so completely insignificant to change the result." is false.

Feel free to provide a link to reliable evidence for this claim that does not refer to material handicaps. Because of your howler with ratings above I'll assume you don't understand anything in the absence of evidence for it.

Again you are putting things into my mouth that I didn't say and claiming I am "wrong" when you have misrepresented my comments. I did not say 3.5/10 would be a bad score. I just said it is not evidence for your claims to be over 2600 strength, especially not in the absence of evidence of benchmarking of the rating differential for the handicap and contempt arrangement.

I played a very large number of games against Stockfish 4 between 2014-5 at even time. I do not make any claims about my playing strength based on these games. I am too busy at present to start on a similar project with the latest versions but I may do so in the future. I have no interest in playing a computer at any form of odds. If someone wants to present odds games as evidence of playing strength they need to play them under controlled observed conditions and benchmark them.

Then beat strong human players in an OTB tournament. If you are as good as you claim to be (i) it will be easy (ii) it will not take much time (iii) it will make you a lot more money than trying to flog your book online to people who can see you are a troll who makes untestable claims.

Irrelevant to my point that there are no examples of merely strong club players proving that they have become very strong just by playing against computers.
Just a brief remark, as you are obviously troll-modding.

The elo you calculated is for Komodo-handicap versus humans-no-handicap.
The real performance of Komodo and the humans can be calculated, when you take this into account.

Well, what to say more, this is not a howler, but a howler to the power of n.

Later

5. Originally Posted by LyudmilTsvetkov
Just a brief remark, as you are obviously troll-modding.

The elo you calculated is for Komodo-handicap versus humans-no-handicap.
The real performance of Komodo and the humans can be calculated, when you take this into account.

Well, what to say more, this is not a howler, but a howler to the power of n.

Later
Mr Tsvetkov, if you are so good, why do not you start playing chess tournaments and make some money this way as well as prove how good you are?

6. Originally Posted by LyudmilTsvetkov
Just a brief remark, as you are obviously troll-modding.
Actually I'm not modding you yet, but that will probably happen soon. You've had more than enough chances to justify your claims to be a 2600+ player and have failed.

The elo you calculated is for Komodo-handicap versus humans-no-handicap.
The real performance of Komodo and the humans can be calculated, when you take this into account.

Well, what to say more, this is not a howler, but a howler to the power of n.
Not my fault if you meant something completely different to what you wrote but didn't explain yourself properly.

7. Let me add more contexts... I'm an experienced chess engine programmer. While I don't want to name myself, I'll share you who "Lyudmil Tsvetkov" is.

This guy, Lyudmil Tsvetkov has been pissing me off for years. He had the encourage to challenge myself and all the top computer chess programmers with his "theories". He wanted us to modify our engine for his computer chess "understanding". For example, he'd sent me a request something like:

>> Lyudmil Tsvetkov: "You are 100% wrong. You don't need engine search, they are useless. Perfect evaluations make good chess. No search. Your engine will gain 1000+ ELO if you follow me."

That was rubbish. He was asking me to remove alpha-beta, and play chess purely by static evaluations.

Lyudmil Tsvetkov doesn't know anything about computer chess, and he's not a strong player. He thinks he's the authority in computer chess, but if he's the authority I'd be the god. Please note he talks a lot, but he has never mentioned anything about chess engine search, because he doesn't know anything about it. Ask him about late move reduction, he'll avoid your question. Ask him to code a simple alpha-beta algorithm, he won't be able to tell you what that is.

Lyudmil is a big fan for computer chess, he really like it. That's his hobby, passion and his "job". Computer chess is everything he has. He doesn't do anything other than computer chess. Lyudmil likes making friends with chess programmers, but we don't like him. He'd requested to add my Facebook, but I refused.

8. Lyudmil Tsvetkov was about FIDE 1900 many years ago when he was still playing competitive chess. Now, I think he's about FIDE 1600-1700.

9. Originally Posted by studentt
Let me add more contexts... I'm an experienced chess engine programmer. While I don't want to name myself, I'll share you who "Lyudmil Tsvetkov" is.

This guy, Lyudmil Tsvetkov has been pissing me off for years. He had the encourage to challenge myself and all the top computer chess programmers with his "theories". He wanted us to modify our engine for his computer chess "understanding". For example, he'd sent me a request something like:

>> Lyudmil Tsvetkov: "You are 100% wrong. You don't need engine search, they are useless. Perfect evaluations make good chess. No search. Your engine will gain 1000+ ELO if you follow me."

That was rubbish. He was asking me to remove alpha-beta, and play chess purely by static evaluations.

Lyudmil Tsvetkov doesn't know anything about computer chess, and he's not a strong player. He thinks he's the authority in computer chess, but if he's the authority I'd be the god. Please note he talks a lot, but he has never mentioned anything about chess engine search, because he doesn't know anything about it. Ask him about late move reduction, he'll avoid your question. Ask him to code a simple alpha-beta algorithm, he won't be able to tell you what that is.

Lyudmil is a big fan for computer chess, he really like it. That's his hobby, passion and his "job". Computer chess is everything he has. He doesn't do anything other than computer chess. Lyudmil likes making friends with chess programmers, but we don't like him. He'd requested to add my Facebook, but I refused.
Among the obvious lies about Facebook friending and requests(I did not even have Facebook account at that time), you say one very important truth:
"He does not do anything else than computer chess."
That should imply I am very strong then, and know my subject perfectly, right, so you are simply contradicting yourself and further rasing
my profile.

Thank you.

10. Originally Posted by studentt
Lyudmil Tsvetkov was about FIDE 1900 many years ago when he was still playing competitive chess. Now, I think he's about FIDE 1600-1700.
That has been already checked and links provided, so try to read the whole thread first.
Otherwise, you sound hollow.

11. Originally Posted by Kevin Bonham
Actually I'm not modding you yet, but that will probably happen soon. You've had more than enough chances to justify your claims to be a 2600+ player and have failed.

Not my fault if you meant something completely different to what you wrote but didn't explain yourself properly.
In view of the imminent threats, I will most probably have to retreat from here at some point.
Before doing that,
I just want to bring back to life an intriguing thread on talkchess, involving a live chess game between me and Stockfish, played in late 2014:
http://www.talkchess.com/forum/viewt...f8debe27a1a99c
This is just to show how much analytical effort has gone into developing the right strategies to overpower the top engines.
With each move consistently analysed for half an hour, and Stockfish using 16 threads, the amount of knowledge one gets from similar sessions is certainly tremendous.

And that is only one of maybe more than a thousand similar analytical threads on talkchess during the last 5 years.

Some might try to raise cheating allegations against me, but I am worth
what I am worth.

12. And please note, the game was LIVE, so I could not have been fixing pgns, TC, etc.

13. Originally Posted by MichaelBaron
Mr Tsvetkov, if you are so good, why do not you start playing chess tournaments and make some money this way as well as prove how good you are?
Actually, a good idea.

14. Originally Posted by LyudmilTsvetkov
In view of the imminent threats, I will most probably have to retreat from here at some point.
Oh don't stop now, this is going so well!

15. Originally Posted by LyudmilTsvetkov
"He does not do anything else than computer chess."
That should imply I am very strong then, and know my subject perfectly, right, so you are simply contradicting yourself and further rasing
my profile.

Thank you.
More nonsense. There are people who are obsessed with chess and play it all the time who are weak players. One of our posters here is always telling us about them!