Well, prove what is a claim and what not.
> All the talk all over the web about equal or level playing fields is missing the point.
@a_pleasant_illusion This is a simple consequence of the fact that chess people are in general interested in chess but not in AI problems. And the question "what engine is the strongest" is important for the chess world as engines are actively used by people, a lot believe that SF actually is, this is why the claims like "AlphaZero demolished Stockfish" attract a lot of controversy.
The point of Deep Mind of course is that they invented a ridiculously effective algorithm to learn machines from total zero to incredible level, albeit in quite restricted but still pretty general set of fields (closed, easy-to-simulate environments with a well-defined and easy-to-calculate performance function). And its incredibility is not decreased by such trivialities as SF not having bases or being of the latest version - 64CPU SF8 is a godlike player even without all this things. That this algorithm can be simulated only on the superhardware - is inferior; the point is that it CAN be simulated at all, and this may change the ACTUAL world, not the chess world. In a sense it is sad that people prefer to discuss Stockfish thing instead, but this is also quite expected in the chess community. Singularitarians obviously discuss the opposite.
@Wolfram_EP Why do you think that 64CPU SF8 with minute per move is a godlike player!? What do you say about 64 000 CPU with 24 hour per move!?
And this is achievable on modern supercomputers like Lomonosov or Titan or other
SF dev on 64000 CPU doesn't make sense, the devs didn't optimize it for such a big number of cores. SMP is not tested very well in fishtest even for 64 cores.
Well, I mean the best machine learning achievement in chess was Giraffe which is hardly a GM level on a reasonable hardware. SF8 is the result of long and hard development with a lot of clever tweaks and tricks and is far beyond a GM level on the smartphone.
Ok, we'll wait for a really good match with more or less equal iron conditions. I hope AlphaZero still will be interested in this, they got a good intermediate result. And someday we will see a similar match.
Now Houdini and Komodo seems to be in best shape.
Houdini just won TCEC tournament
> I hope AlphaZero still will be interested in this
@GuardianRM Deep Mind is not interested in creating a computer chess world champion, it is interesting in creating an Artificial General Intelligence. Chess and Shogi are more like "well, these seem to be easy problems, let's spend a couple of months to them and move to the next milestone".
I think they could create a machine that dominates the current top engine trinity in a more convincing way but what's the point for them? Considering the enourmous amount of resources they used for this learning, it will be extra hard to obtain big money on this, and in the long-term prospective chess is useless for Google, unlike general AI algorithms. The latter ones are the only reason why they are actually using these resources.
@ilidiomartins This is arguable. TCEC is not a reliable benchmark due to quite small amount of games played (yes, even 100 superfinal games is a small number in computer chess). If you look at Stage 2 results, you'll notice that the score was 18.5-18.5-18.0 H-K-SF respectively, and the fact SF didn't make it can as well be unluck. CCRL list (http://www.computerchess.org.uk/ccrl/4040/ ) is more reliable due to a bigger number of games. All in all, at the moment the difference in skills of H, K and SF is quite minor and who is superior often depends on secondary conditions like time control, hardware, presence of endgame tablebases etc.