dear, dear a tiny down tick

[main line]

Did you notice 300Elo improvement in last 400.000 games? We need enthusiasts like before when we made 400.000 games daily.
It means we can make improvement +300 Elo daily. Probably Stockfish will be strongest engine next 2-3 weeks only. No more.

[shrapnel]

Probably Stockfish will be strongest engine next 2-3 weeks only. No more.

Now, THAT'S the Spirit.
LCZero's rapid progress has got the old alpha-beta engine supporters worried.
Remember these are the very same people who simply couldn't believe it when AlphaZero DEMOLISHED Stockfish and kept spouting ridiculous Conspiracy theories.
Unfortunately, Google's lack of interest in Commercializing AlphaZero has given a fillip to the nay-sayers.
Now, they are a little bit worried about LCZero.

[stavros]

the fanatism around chess engines is outstanding ,surpassing what is called "a board game" or "hobby"
from rybka era most ppl care to watch their favorite engine to crush the opp. engine,just like football fanatism
wondering if its the top priority in their life...so scary
meanwhile the chess as hobby is dying front of the modern "arena" full of maniacs watching robots to fight and instead of blood to
watch fragments of microchips.
so now the menu includes stockfish vs lc0, who will be worried? and who will be happy?
the meaning of the google A0 project was to show via a board hame (chess) that AI nowadays can help really to humanity..
i know i know..humans now will take this revolution to make what else? smarter weapons first.
the science like a slave, under interests,weapons,money,and people control.

[Dann Corbit]

Did you notice 300Elo improvement in last 400.000 games? We need enthusiasts like before when we made 400.000 games daily.
It means we can make improvement +300 Elo daily. Probably Stockfish will be strongest engine next 2-3 weeks only. No more.

Having just lost 200 Elo, that is a net of +100.
300 Elo per day won't hold, but I do expect that we will see big, long term gains again. Unless there are more bad patches.

[Dann Corbit]

{snip}
i know i know..humans now will take this revolution to make what else? smarter weapons first.
the science like a slave, under interests,weapons,money,and people control.

Humans do have an unfortunate tendency to put shiny, new technology to the worst imaginable use first.
But it is not the fault of computer chess.

[Henk]

This is only 'simpel' reinforcement learning where it can generate training examples easily.

This is not the case in other situations where problem is acquiring enough training examples.

By the way also tremendous amount of computing power was needed.

Also research has been done by deepmind.

[Milos]

I know my error bars are quite big. Still they are better than yours, which are non-existant.

I have my own results and I really don't feel obliged to post it to you, plus I see results that other ppl post, much more reliable results, so I don't need yours which are only marginally better than that self-play BS curve that fanboys adore, especial since on more than one occasion before they've proven to be faulty.

[Henk]

There progress is 1 of their Elo points per 12000 games. So maybe that's 0.5 normal Elo point per 12000 games.
So for 300 normal Elo extra they need to play 600 * 12000 = 7.2 million games.

If long term trend doesn't get broken.

If 150000 games are played per day then we have to wait 480 days. Say one and a half year.
Have fun waiting (not)

[yanquis1972]

There progress is 1 of their Elo points per 12000 games. So maybe that's 0.5 normal Elo point per 12000 games.
So for 300 normal Elo extra they need to play 600 * 12000 = 7.2 million games.

If long term trend doesn't get broken.

If 150000 games are played per day then we have to wait 480 days. Say one and a half year.
Have fun waiting (not)

what's your basis for that elo estimation? i don't believe there's a universally acknowledged reliable source so i'm curious

edit: nvm, apparently it's the selfplay graph & then arbitrarily dividing in half

[Henk]

Yes I'm too stupid to do complex calculations. So I divided their Elo in half (divided by 2). It's only very crude estimations.