Re: XieXie Master -- a recommendation

On Mar 5, 12:31 am, mikeh...@[EMAIL PROTECTED]
 wrote:
> On Mar 5, 12:09 am, mark <mschr...@[EMAIL PROTECTED]
> wrote:
>
> rybka is certainly < 3100 at 120/40

Look at the CEGT 40/120 rating for Rybka on a 4CPU after 500 games at
www.husvankempen.de/nunn/40_120_ratinglist/Quad/qratinglist.html

> 1.  discounting his oversight of the mate-in-1, the final score
> would've likely been 3.5-2.5.  given the match was only 6 games long,
> i'd hardly say that was a conclusive result.

The score or number of games does not make the result conclusive but
the opponent and the conditions. Kramnik played a weak program on a
slow computer while getting help from the computer and he still lost.
That was in 2006 and the computer's rating only go up with time.

> 2.  kramnik had no incentive to either win nor play for the draw.  he
> was guaranteed the prize money regardless of the outcome.

Not true. For winning the match Kramnik would have won an extra
500,000 Euros.

> >I think that today in a fair tournament the world champion would lose
most of the games and draw maybe 10 percent of the games.
>
> first of all, are u talking about a tournament (i.e. round-robin
> format), or a match (e.g. human vs. computer)?  i'll presume u mean
> the latter.  what are the time controls?  if they're 120/40, no way.

Either one it does not matter. I meant a match at 40 moves / 120
minutes. Rybka is over 300 points stronger than Kramnik on today's 16
cpu computer.

> check out the wiki article on shogi and u'll see u're wrong.  a
> computer will beat the world's #1 way before 2027.  wouldnt be
> surprised if it happens by 2010.

Which wiki article on shogi? Do you mean the Wikipedia encyclopedia
article on shogi? They talk about the game between Watanabe and
Bonanza. Watanabe did not rate the computer highly.

> schaeffer doesnt say what u claim he says in this
article:http://spectrum.ieee.org/jul07/5379

I think you are right about strongly solving the 3 move ballots. I
reread the definition of strongly solved and what Schaeffer is doing.
He will weakly solved checkers because he stops searching whenever he
finds a win.

> btw, i've done tests of neuchess playing itself.  red's move are
> evaluated to at least 17-ply for every move, while the black side only
> gets up to 14-ply.  this setup simulates an overclocked octal system
> running neuchess against a dual-core system at time controls of 80min/
> 30s per side.  i choose the opening by which black selects against
> red's opening selection.  this occurs for the first 10 moves or so.
> after that, i let the engine play itself.  so far every game has ended
> in a draw.  this is why i believe Xu can do the same regardless of how
> powerful the hardware is that neuchess is running on.

That's counterintuitive. Why doesn't the 17 ply set a trap that the 14
ply won't see because of the horizon effect? The 17 ply should win.