What is Lc0-dev, now playing at Chess.com...

[jimgl...@gmail.com]

... against Stockfish, Xiphos and Lc0?

[Ingo Weidner]

This seems tobe a demo tournamnet before CCC 5 starts.

In the description they mention ID 40564 for "Lc0-dev". You get the description by clicking at the engine name in "Standings"

The "normal" Lc0 engine there seems to use ID 32890..

[jimgl...@gmail.com]

Thanks.

[LuckyDay]

lc0-dev is essentially using the most recent release candidate lc0 engine, and grabs whatever the latest T40 network is at the time (it actually has a script to do so, so it is constantly updating its T40 net)

[jimgl...@gmail.com]

So far Lc0-dev is doing pretty well, the only entry with a winning record at this moment in the "escalation" format. See the crosstable.

https://www.chess.com/computer-chess-championship

None of the top seeded entries have played yet, but Lc0-dev looks like at least being at the top of the second division. Pretty good being she's still just a baby, and how the 'weaker' engines she's beating are in the top 10 in the world.

[Vassilis]

Just amazing...

Nice idea to always choose the latest net.
Those t40 nets look really promising.
Only loss to Xiphos, by now.

[jimgl...@gmail.com]

Lc0-dev is winning its match against Ethereal. Very impressive, IMHO.

[jimgl...@gmail.com]

Lc0-dev -- Test 40 -- is rockin' it at Chess.com, dominating the escalation.

She's 20.5/33, so far up that no other engine has a winning record. She just beat Fire in their first game.

The big four (including adult Leela test 30) haven't played yet, but she's looking like she may be #5 already, while barely a toddler.

I'm looking forward to TCEC 15 already.

[Moronocracy_USA]

CCC mentions that for each game the last net of BBLL(c)  = babyLeela = Lc0 dev is used. 

If BBLL(c) would have played all her games last, her results would even be better than they already are !!!!

It will be interesting to see BBLL(c) playing against the top 4 - with each time a better net ;-)

Op zaterdag 2 februari 2019 15:22:50 UTC+1 schreef jimgl...@gmail.com:

[jimgl...@gmail.com]

Test 40 is now 3/4 against Fire, the #5 seed.

BabyLeela40 is crushing it.

[Michael Elkin]

The network is probably somewhere in 4061x. It seems to be getting much stronger the longer this goes. At the way it is going I would not be surprised if it ends up beating all engines, even T30 and SF.

[Moronocracy_USA]

BBLL(c) = babyLeela v Fire: +3 -0 =2   !!!!!!

Op zaterdag 2 februari 2019 18:06:33 UTC+1 schreef Michael Elkin:

[Moronocracy_USA]

BBLL(c) = babyLeela v Fire: +4 -0 =2   !!!!!!


Op zaterdag 2 februari 2019 18:13:46 UTC+1 schreef Moronocracy_USA:

[jimgl...@gmail.com]

Leela beats Fire *again*, now 5/6!

And she did it outplaying Fire in an even ending.

*Until* she had a mate-in-3 up two Queens and a Bishop versus naked King - after which she of course sacrificed both a Queen and the Bishop to mate in 49!

Go Test 40 Go!

[jimgl...@gmail.com]

BabyLeela40 so totally crushed Fire (6.5/8) that I thought there might have been something wrong with Fire's setup, technically.

But Fire just pounded Xiphos (which gave Leela her only loss) by 5.5 - 2.5, so I guess not.

It makes Baby's performance look even more credible. I'm looking forward to seeing this baby grow up.

[Ingo Weidner]

Lc0-dev just had a win and a draw against Komodo (the "big" K, not MCTS). I admit i did not expect that.

After my current test of new 32xxx networks is finished it will be finally time to test the latest 4xxx nets. I already downloaded some of the latest ones (40600 and above).

[Moronocracy_USA]

BBLL(c) vs Komodo 2255 (newer version than in TCEC Cup): 4.5 - 3.5 !!!!  (+1 -0 =7)

BBLL(c) beats Komodo as well - The winning streak continues !!!!

Upcoming (game #129 now playing):

#169 BBLL(c) vs Houdini

#225 will feature Leela vs BBLL(c)

#289 BBLL(c) vs SF 010219

Newer BBLL(c) nets are used next game after being dropped. So the Top 3 will face increasing resistance from BBLL(c) ;-)

Op zondag 3 februari 2019 00:01:47 UTC+1 schreef jimgl...@gmail.com: