Toby-Tal update

    My name is Tim Tobiason and I am the co-author and owner of
TobyChess and the distributor of the TobyTal chess engine. In 2003, I
hired a programmer to create Deep Tactics, an interface which does many
wonderful things. Early on my programmer wrote a chess engine which
found every checkmate at the end of every game in the chessbase database
which we stored. It then analyzed backwards finding every possible way
to force mate. We stored these and that is how we created the 12 million
checkmates which we sell with Deep Tactics and Toby Tal. We discovered
that the average mate in 4 has 140 solutions and the average mate in 6
has over 1,000 ways to mate so we have a reader which displays every
branch of the tree in our trainer. We called this engine nutcracker and
it is part of TobyTal.

    A year ago we created a new engine called serpent which beat Rybka
every single game. It beat every engine every time except Hiarcs which
got 4 draws in 16 games. Its elo rater rating was 3600+. It was designed
to run on 128 bit machines and we ran it off of a server. This engine
did not use trees, nodes, branches, etc. It created a new kind of
table-base it could make on the fly and then plan on how to achieve the
future positions. Authorities stepped in and stopped us from selling it
and no amount of money would have changed this. I cannot divulge the
rest of the story at this time.

    I asked the programmer over the last year to make a new engine with
our best ideas and last November he provided me with a test copy of
Toby-Tal. It was small, 120K. It used nutcracker and ideas from Serpent
although we still use searches, nodes and trees. I showed this to a
customer in Kansas at that time and was told about a new engine that was
out called Robollito. We did not release TobyTal at this time because we
had a serious hash table read error which caused us to switch from 32
bit hash tables to 64 bit hash tables. This fixed the problem of our
engine getting weaker over time from read errors from the old hash tables.

    On release of our engine a couple of weeks ago by e-mail and at the
North American Open, I found I had received profane e-mails from
"anonymous" accusing me of selling a clone (Robo). My website was
attacked with millions of hits and taken out of business along with my
e-mail for about a week while I was on the road. We are now back up
running and I will address the issues which have been raised.

1. TobyTal is a unique engine which we created. It is 120K in size. Robo
is 301K and Rybka about 2.3mg. Our engine is encrypted but not
compressed. This is to protect our code which I paid to create and which
I personally contributed to. It is our legal property with dozens of new
ideas which I intend to protect. This is the first engine in which part
of its program was actually written by another program.

2. It solves about 1/2 of the Nolot puzzles on Wikipedia. These are
puzzles too hard for chess engines to solve. On puzzle 9 we found a
novel solution (Nxh6) which wins in every variation! Neither Robo or
Ryka can solve more than 1-2 of the 20 puzzles. We find unique moves in
many situations because of a special King attack routine we invented. I
have seen arguments accusing us of being the same as Roby, etc because
we play the same moves in a position that they do. When I run the
positions in Fritz 12 it also finds the same moves, therefore Fritz 12
must be a clone too?
3. We have multi move and search move and Robo does not. We do not have
tablebase support yet. We also have nutcracker and serpent code inside.

4. We do not run on most windows XP systems (which we hope to fix soon)
while the other engines do.

5. Using a checksum to compare the engines, we found about 5% identical
code with Robollito which I consider convenience code used by all chess
programs. Things like Quintz search and null move are common among all
strong programs and we use some of these as wells.

6. We play better chess than Robo or Rybka by 20-80 points right now
which is a lot at that level. If you look at the games, they run in
streaks. When we lose in the streaks, we have a hash table read error or
other major bug which only happens in play game mode. It does not happen
in analysis mode. Once this bug is found and fixed we may play a hundred
points or more above Robo & Rybka!! 

7. Our engine is 32 bit bit but plays well vs the 64 bit versions of
Robo and Rybka.

7. In the end if the Robo people wish to resolve this in court, we will
happily submit our code to the court for 3rd party comparison under the
courts supervision to ensure our code remains secret.

    I am selling our engine on-line and at tournaments. At the North
American Open in Las Vegas, we sold 21 TobyTal's while Chess Palace told
me they sold 5 Fritz 12's and 4 Rybka's. People could see the engine
matches and install them on their own computer and test them themselves.

    I do not sell the engine online as a download. We only sell it with
the Deep tactics interface and to prevent piracy, future versions will
be tied to the interface like Fritz 12.

    The issues will be settled on the chessboard in the coming months.
We will add tablebase support, new chess knowledge and make a windows XP
version in the next few weeks and will hopefully fix the bug in the
current version. I will send this to all of you as a free upgrade if you
have already purchased TobyTal [Thanks for the business]. After that we
will create multi core and 64 bit versions with opening learning and
multi computer capabilities. This will be sold as Deep TobyTal for $89
or an additional $30 upgrade.

    In 4-6 years, there ill be no more chess engines as chess will be
solved and the moves will simply be looked up in tables (similar to the
dinosaur methods of tablebases used today). Then all these chess engine
arguments will be obsolete.

Toby