BubbaTough wrote:Instead of adding 200 ELO, I would find it more interesting to add 90 minutes to clock.
I told it to lkaufman at RF, and he replied:
"Yes, I think that would be interesting too. I could probably persuade one of our local players rated around 2200 FIDE to play a couple of tournament level games here at knight odds for a nominal prize."
perhaps 90 more minutes is worth more to the human than 200 extra elo strength. Or perhaps less.
But I think that both together would be more to the point.
Ovyron wrote:Now he's trying to find a 2400 rated opponent that is willing to play.
He found the equivalent to a FIDE 2435, and here are the results:
lkaufman wrote:Tonite my Rybka (latest version on quad) played a five game knight odds semi-blitz match (5'+5" increment) against a much stronger opponent than the previous such match. His name is Jared Defibaugh, ICC handle "trixr4kids", and his ICC blitz rating is 2706. According to the estimated conversion formula, this corresponds on average to a FIDE 2435 player, which would be an above-average IM. His actual FIDE rating is only 2080, but he's only 18 years old and that rating does not reflect his current strength. Probably his real strength at a time limit like this is typical of a 2300 FIDE player, i.e. FM level.
The result was 3 1/2 for Rybka and 1 1/2 for Defibaugh. It could have been closer had he claimed a draw by three-time repetition in the first game, but perhaps not fully realizing the strength of Rybka he played for the win and lost. Rybka played a different first move in each of the five games. After the match, Defibaugh commented that Rybka seemed unlike any program he had played before, both in strength and style. He felt he could have done better at a longer time limit, which of course is probably true.
I'll post the games below.
Rybka repeated despite the maximum contempt factor setting, as she was still down the Exchange. I was hoping she would play for the win, as the position is not as bad as the initial one.
OK, very well then!
I do find this impressive for the human, even though it's knight odds.
At this rate, longer time controls might easily make him equal or even higher. I would think he should get 3.5-1.5 to him.
This STILL gives a high estimation of Rybka without a knight, higher than what others might think. However, even at this, I would suggest that maybe this player is an exceptional computer player, and that not all at his elo would do as well. Maybe older people (with identical rating) would do less good?
At any rate, I personally think it is an interesting challenge to find if humankind CAN get perfect results vs top computers at knight odds.
I mean, shouldn't it always be a won game from the start?
I love the presentation of a FIDE 2080 player as a 2435 IM. There is no way the player can officially claim this 400 point jump! I know many players who specialize in blitz and do not have a decent FIDE rating. Many ICC players will magically start becoming IMs based on their blitz skill if this conversion formula was used. The problem for many of these players is that when both players get more time, their blitz advantages reduce. Of course, I have nothing personal against the person mentioned by Larry Kaufman.
That being said, the time control probably makes it harder to beat Rybka even when a piece up.
I agree with shiv. The gap in strategic understanding between a FIDE 2080 and an IM would certainly impact the ability to exploit this advantage against a computer.
shiv wrote:I love the presentation of a FIDE 2080 player as a 2435 IM. There is no way the player can officially claim this 400 point jump! I know many players who specialize in blitz and do not have a decent FIDE rating. Many ICC players will magically start becoming IMs based on their blitz skill if this conversion formula was used. The problem for many of these players is that when both players get more time, their blitz advantages reduce. Of course, I have nothing personal against the person mentioned by Larry Kaufman.
That being said, the time control probably makes it harder to beat Rybka even when a piece up.
Fide rating can be misleading.
People can clearly be at level of 2400 and have lower fide rating if they did not play in the last years in human tournament or if they startes as 2000 some years ago and improved but did not play in many tournaments for fide rating or if they started only recently and had one bad tournament when they started.
Inspite of it I can say that in this case there is no reason to think that
the fide rating that he has is too low based on his results because in the last tournaments from 2007 that were considered for fide rating he lost fide rating
I reposted your replies at Rybka Forum, and Larry answered:
lkaufman wrote:With respect to the conversion of ICC ratings to FIDE ratings, I only mean to say that (for example) a player with an ICC blitz rating of 2706 probably plays blitz (specifically 3 min. blitz) about as well as the average 2435 IM does, including those IMs who only play blitz occasionally. Since these games were actually more like 10 min. games, the player's effective rating is somewhere in between his tournament rating and his 3 min rating on ICC. There is no doubt that the longer the time limit, the better humans will do with a handicap; the question is how sharply do these results rise with time.
Permanent Brain wrote:If someone could play blitz at 2435 level, wouldn't he at least have 2300+ FIDE, USCF or USCF-Quick ratings?
I believe that rating conversions from server to federation ratings are not reliable (if not plain wrong, often).
lkaufman wrote:Usually that would be so, but there are some people, mostly young players, who play countless thousands of blitz/bullet games on the internet but who have played only a small number of tournament-level games. Twenty or more years ago the correlation between blitz skill and tournament level skill was pretty high, because there was no opportunity to play 20,000 blitz games in a year or two. So now we see some extreme differences between blitz and slow skill. But the server ratings are a bit random, because the 16 point gain/loss per game with equal player is way too high, and because players can choose their own opponents. That's why, at least on ICC, the "5 minute" ratings are far more reliable indications of strength than "blitz" ratings, because in the 5 min pool you must play whomever the server assigns for you to play. Still, you should look at someone's average rating, not his rating at one instant of time.
Your beliefs create your reality, so be careful what you wish for.
