The next revolution in computer chess?

[corres]

...
NNUE engines don't need expensive GPU cards.

This is the only one real benefit.

Sure, everybody is wrong and you are right.

The future will decide about who is wrong and who is right.
In every cases I keep my NVIDIA cards.

Not sure if a such a thing happened ever before, #1 top entry on its first run @CCRL...
better invest in some AVX based CPU.

Some ten Elo plus is not the World, especially if you use AVX based CPU.
After some months NVIDIA series 3000 will come and the next years the series 4000...
As if you made some post about it...
I think NNUE is an interesting trying but not a mark of a new paradigm. This was the AlphaZero only.

[Ovyron]

Uly, you are simply wrong. This project is at its beginning.

But... that's what I said.

Can you imagine when NNUE gets as much training as Leela has today, how will it play?

I can't, the string where Sergio was adding 4 elo every 5 hours broke my brain.

[chrisw]

...
NNUE engines don't need expensive GPU cards.

This is the only one real benefit.

Sure, everybody is wrong and you are right.

The future will decide about who is wrong and who is right.
In every cases I keep my NVIDIA cards.

Hehe, dude, the future is now

http://ccrl.chessdom.com/ccrl/404/

Not sure if a such a thing happened ever before, #1 top entry on its first run @CCRL...

better invest in some AVX based CPU too

--
Srdja

Well, I am going down the contrarian path. NNs have their own exploitable problems too. For the time being.

I mentioned it before, I guess this will be some kind of co-evolution between
expert-systems, handcrafted eval, and NNs. We see this in other domains too.

We will see which approach will perform better in the long run. But the paradigm
shift is already here, it came stepwise, NN eval vs. expert-system, if we take
a look closer, then the first steps were already the automated tuning methods of
our evaluation functions, NNs are just the next step in this line.

Further, if we consider that Reinforcement Learning offers us new paths, then I
would even go that far to call this beyond super-human-level chess, I would call
this RL ANN based eval "Trans-Human-Chess", cos it comes up with lines which
humans do not come up with, neither the machines they use to program by hand,
or alike, imho.

--
Srdja

Your argument depends on the unproven assertion, because nobody actually really ever tried, because they never needed to (changed now of course), to get the necessary non-linearities into the hand crafted (and then tuned) eval function. NNs get the non-linearities in without human ingenuity having worked out how, it magically dropped out via the known engineering by technical field (chess) ignoramusses, ignorami, whatever.
NNs are entirely dependent on blundering their way to whatever one of the multiple partial optima are in the weights. Nobody knows why or how. We’ll see if smart humanity can get past lucky shot NNs in whatever hardware/software combination. Well, maybe we won’t, it’ll depend on whether anyone can be bothered or not.

[corres]

...
I mentioned it before, I guess this will be some kind of co-evolution between
expert-systems, handcrafted eval, and NNs. We see this in other domains too.
We will see which approach will perform better in the long run. But the paradigm
shift is already here, it came stepwise, NN eval vs. expert-system, if we take
a look closer, then the first steps were already the automated tuning methods of
our evaluation functions, NNs are just the next step in this line.
Further, if we consider that Reinforcement Learning offers us new paths, then I
would even go that far to call this beyond super-human-level chess, I would call
this RL ANN based eval "Trans-Human-Chess", cos it comes up with lines which
humans do not come up with, neither the machines they use to program by hand,
or alike, imho.

Yes, we will see which approach perform better, but no any paradigm shift, only continuous development.
Reinforcement Learning is only a tool to enhance the "contrast" in NN images, but this tool can not give missing information to the NN.

[smatovic]

...
NNUE engines don't need expensive GPU cards.

This is the only one real benefit.

Sure, everybody is wrong and you are right.

The future will decide about who is wrong and who is right.
In every cases I keep my NVIDIA cards.

Not sure if a such a thing happened ever before, #1 top entry on its first run @CCRL...
better invest in some AVX based CPU.

Some ten Elo plus is not the World, especially if you use AVX based CPU.
After some months NVIDIA series 3000 will come and the next years the series 4000...
As if you made some post about it...
I think NNUE is an interesting trying but not a mark of a new paradigm. This was the AlphaZero only.

Yea, "ten Elo plus" with its "first shot"

Looking forward to the LC0 vs. SF NNUE vs. SF battles in here

Agree, Reinforcement Learning with A0 in Go was the real big deal.

But I have to mention your bad quoting practice in here, better not to edit an quote, leads to misunderstandings etc.

...
better invest in some AVX based CPU too

--
Srdja

[smatovic]

Well, I am going down the contrarian path. NNs have their own exploitable problems too. For the time being.

I mentioned it before, I guess this will be some kind of co-evolution between
expert-systems, handcrafted eval, and NNs. We see this in other domains too.

We will see which approach will perform better in the long run. But the paradigm
shift is already here, it came stepwise, NN eval vs. expert-system, if we take
a look closer, then the first steps were already the automated tuning methods of
our evaluation functions, NNs are just the next step in this line.

Further, if we consider that Reinforcement Learning offers us new paths, then I
would even go that far to call this beyond super-human-level chess, I would call
this RL ANN based eval "Trans-Human-Chess", cos it comes up with lines which
humans do not come up with, neither the machines they use to program by hand,
or alike, imho.

--
Srdja

Your argument depends on the unproven assertion, because nobody actually really ever tried, because they never needed to (changed now of course), to get the necessary non-linearities into the hand crafted (and then tuned) eval function. NNs get the non-linearities in without human ingenuity having worked out how, it magically dropped out via the known engineering by technical field (chess) ignoramusses, ignorami, whatever.
NNs are entirely dependent on blundering their way to whatever one of the multiple partial optima are in the weights. Nobody knows why or how. We’ll see if smart humanity can get past lucky shot NNs in whatever hardware/software combination. Well, maybe we won’t, it’ll depend on whether anyone can be bothered or not.

I guess these questions are beyond my knowledge/domain, are we with our predicate logic kind of thinking able to grasp the non-linearities encoded in NNs via RL? Dunno

--
Srdja

[chrisw]

Well, I am going down the contrarian path. NNs have their own exploitable problems too. For the time being.

I mentioned it before, I guess this will be some kind of co-evolution between
expert-systems, handcrafted eval, and NNs. We see this in other domains too.

We will see which approach will perform better in the long run. But the paradigm
shift is already here, it came stepwise, NN eval vs. expert-system, if we take
a look closer, then the first steps were already the automated tuning methods of
our evaluation functions, NNs are just the next step in this line.

Further, if we consider that Reinforcement Learning offers us new paths, then I
would even go that far to call this beyond super-human-level chess, I would call
this RL ANN based eval "Trans-Human-Chess", cos it comes up with lines which
humans do not come up with, neither the machines they use to program by hand,
or alike, imho.

--
Srdja

Your argument depends on the unproven assertion, because nobody actually really ever tried, because they never needed to (changed now of course), to get the necessary non-linearities into the hand crafted (and then tuned) eval function. NNs get the non-linearities in without human ingenuity having worked out how, it magically dropped out via the known engineering by technical field (chess) ignoramusses, ignorami, whatever.
NNs are entirely dependent on blundering their way to whatever one of the multiple partial optima are in the weights. Nobody knows why or how. We’ll see if smart humanity can get past lucky shot NNs in whatever hardware/software combination. Well, maybe we won’t, it’ll depend on whether anyone can be bothered or not.

I guess these questions are beyond my knowledge/domain, are we with our predicate logic kind of thinking able to grasp the non-linearities encoded in NNs via RL? Dunno

--
Srdja

Yes, else ply 0 Leela would beat GMs at non-blitz and GMs/IMs can mostly express in language why so and so move and so on.

The main problem for getting in the non-linearities is already solved by tuning. The difference between tuning a polynomial of known themes and a layered spreadsheet of one hots, is that human can understand the polynomial and devise more terms and combine terms, all on basis of human ingenuity, whereas the layered spreadsheet technician has absolutely no idea at all what is going on (well maybe some heat maps, but those are in effect no more than pretty pictures). Spreadsheet technician is at the mercy of multi-dimensional connections that make no readable sense.

[corres]

Uly, you are simply wrong. This project is at its beginning.

But... that's what I said.
Can you imagine when NNUE gets as much training as Leela has today, how will it play?
I can't, the string where Sergio was adding 4 elo every 5 hours broke my brain.

As the chess power of default Stockfish(dev) and Stockfish NNUE is near the same, it proves the very small NN for Stockfish NNUE rather well represent the evaluation of Stockfish moreover the evaluation of Stockfish is very simple comparing to itself to the chess play we can imagine only a much smaller development in NN for NNUE than in NN for Leela.

[Ovyron]

the chess power of default Stockfish(dev) and Stockfish NNUE is near the same

First it was much weaker. Then it was weaker. Then it was near the same. Then it was stronger. Then it was much stronger. See where this is going?

NNUE showcases the potential of Stockfish's eval improvement, as it has a cost and makes it search slower, but nothing stops Stockfish's eval from being improved and reach that same strength without slowdown. But then the new improved eval can be used for a new NNUE that is again slower but stronger, creating a virtuous cycle.

This is like a baby with 200 IQ and 200 EQ, tremendous potential, but still a baby. But she'll grow.

[chrisw]

the chess power of default Stockfish(dev) and Stockfish NNUE is near the same

First it was much weaker. Then it was weaker. Then it was near the same. Then it was stronger. Then it was much stronger. See where this is going?

Would you like to buy this dotcom business I have to sell, it’s been going up and up and up and up and everyone can see where this is going (up).

NNUE showcases the potential of Stockfish's eval improvement, as it has a cost and makes it search slower, but nothing stops Stockfish's eval from being improved and reach that same strength without slowdown. But then the new improved eval can be used for a new NNUE that is again slower but stronger, creating a virtuous cycle.

This is like a baby with 200 IQ and 200 EQ, tremendous potential, but still a baby. But she'll grow.

Unless it’s an orangutan that is almost fully developed brain wise and won’t learn to read and write however hard you try.