LCZero update

[Milos]

If you think the Stockfish they used was that crippled, then why don't you replicate a match where you measure the strength of Stockfish 8 on 64 threads, 1 GB hash tables, and no book or table bases, against Stockfish 9 with any hardware or tuning you can think of. At that point you would have a discussion basis for the assertion that the Stockfish in their match is crippled, but not before.

The similar matches have already been run resulting in strength difference of more than 100Elo. You should inform yourself.

AlphaZero simply initiated a paradigm shift in computer chess programming, by showing that something completely different works as well or even better with some refinement.

Fanboyism usually doesn't bring much facts. In this particular case, paradigm shift, beside some murky preprint used as an advertising effort by a multi-billion dollar advertising company, is not based on a single real piece of evidence. It is only wishful thinking atm.

[David Xu]

So no, it does not simply have a low draw rate because it plays "like a three-year-old child". Wonder what it will take before you people start to understand.

It certainly takes much more than an arrogant and delusional troll preaching them.

A remarkably concise self-description on your part. Well done.

Fanboyism usually doesn't bring much facts. In this particular case, paradigm shift, beside some murky preprint used as an advertising effort by a multi-billion dollar advertising company, is not based on a single real piece of evidence. It is only wishful thinking atm.

Anything in the range of 3300+ Elo is far in excess of anything that had previously been achieved by reinforcement learning in the past (not least due to lack of interest on the part of the existing computer chess community, many of whom seem to enjoy pretending that innovation is no longer possible). In that sense, it doesn't matter whether AlphaZero is 100 Elo stronger than Stockfish or 100 Elo weaker: both are revolutionary results in the field of machine learning.

Of course, everything I just wrote in the above paragraph is for the benefit of others who may be reading this thread; I have little doubt that you will continue to stick to your outdated views, right up until the day that LCZero defeats Stockfish comprehensively.

[jkiliani]

The similar matches have already been run resulting in strength difference of more than 100Elo. You should inform yourself.

If you know something here, cite a source. 100 Elo sounds plausible to me by the way, but it does not in the least impact my point, which was that if AlphaZero had scored about even against a tuned Stockfish 9 instead of defeating a somewhat mistuned SF 8, that would not diminish Deepmind's achievement in any way.

Fanboyism usually doesn't bring much facts. In this particular case, paradigm shift, beside some murky preprint used as an advertising effort by a multi-billion dollar advertising company, is not based on a single real piece of evidence. It is only wishful thinking atm.

While Deepmind does have a tendency to present their stuff in the most favourable light possible, they aren't in the business of making things up. If they had done that, it would have blown up in their faces long ago, so the game records that they did publish (and were extensively analysed by the community) are actual evidence that their program works.

The majority of the computer chess community might have some reservations about the match conditions with Stockfish, but not about whether Deepmind's achievement was real. If accepting a long list of evidence that reinforcement learning works for two-player complete information games as true instead of sticking my head in the sand makes me a fanboy, so be it.

[Milos]

So no, it does not simply have a low draw rate because it plays "like a three-year-old child". Wonder what it will take before you people start to understand.

It certainly takes much more than an arrogant and delusional troll preaching them.

A remarkably concise self-description on your part. Well done.

That is the second best reply you ever came to.
After all, it has 2 sentences in stark contrast to 18 out of your 20 total posts on this forum that contain only 1 sentence. Deeply profound really.

[David Xu]

I should note that Milos Stanisavljevic is a known troll around these parts, as should have been made fairly clear by his posts already (such as, for instance, the fact that he referred to Google DeepMind as an "advertising company").

Such attention-seeking is best not rewarded.

[Ozymandias]

draw rate is low only because it plays on a level of 3-years old kid.

Experience shows that draw level is not so low at the lowest human level of children and you often get stalemates.

True, the also have a tendency to shake hands, for fear of losing.

[Werewolf]

The similar matches have already been run resulting in strength difference of more than 100Elo.

The whole experiment lacked fairness, which coming from a world-leading organisation is deeply disturbing.

[noobpwnftw]

Oh, I think we gave AlphaZero's attention-seeking too much reward already.

Basically if some guy who had some achievement in chess programming is now saying that he can make an engine that have some fancy tech(or simply use large persistent TT) and then provide you a 100 game match result that it beats SF9 with no book, do you buy it?

There is nothing wrong with fancy tech and it is true that generalized approaches may perform equal or even better, and for no matter what reason should we deny the idea of innovation, but none excuses them from the fact that their report is more of a advertisement than real science, and it's coming from whoever we gave credit in the first place.

[jkiliani]

The similar matches have already been run resulting in strength difference of more than 100Elo.

The whole experiment lacked fairness, which coming from a world-leading organisation is deeply disturbing.

Deepmind is a company and primarily concerned with shareholder value, not fairness. I agree that their match with Stockfish was neither very transparent nor fair, but in spite of these caveats, the impact of those restrictions on SF performance can be estimated. 100 Elo implies that AlphaZero would be about equal to a better tuned Stockfish.

Deepmind tends to be selective about what they publish (their successes only), they tend to present them as favourably as they can, and they don't as a rule give out information they haven't already published in papers, or at least in presentations on conferences. But they do publish a lot of impressive stuff, and on the Go side at least, their concepts have been proven to work, also when reimplemented by others.

[noobpwnftw]

One thing to mention about the NN approach is when it works it works, but when it doesn't, you usually can't fix it. There is an example in recent Go engines recreated from AG paper that would fall for some particular position every single time and the amount of training samples wouldn't be enough correct it, not to mention that doing human intervene contradicts the philosophy of "learning on its own", so maybe this is what they choose not to tell you and we probably will never see any A-series gets published for public use.

Sounds funny when you think of their medical plan, do the AI need to kill enough patients before learning not to, if so, how many, a zillion?