Talking About AlphaZero

It's been three weeks since we learned that AlphaZero Is Back!. When the news first broke, what was the reaction from the community of chess engine developers? Taking a few pointers from a recent post, Talking About Chess Engines (October 2018), I checked with the groups behind Stockfish and Leela. First, here's the Stockfish reaction (groups.google.com/fishcooking):-

• 2018-12-06: "A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play" • 'The PGN files indicate the games were played on 18 Jan 2018 [...] It's great to see the new games, but is this just old news?'
  
  
• 2018-12-08: Re: AlphaZero new paper (Science Dec 2018) • 'Alphazero used 5000 first-gen TPUs to play selfgames and 16 second-gen TPUs to train neural net.'

You might expect a deeper look from the Leela (LC0) crowd, but the reaction was also considerably understated (groups.google.com/lczero):-

• 2018-12-06: New AlphaZero 2018 Papers (Discussion) • 'Based on AlphaZero 2018's 1000 game matchup between Stockfish 8, AZ 2018 is only +52 Elo to Stockfish 8, which means it is weaker than Stockfish 9.'
  
  
• 2018-12-07: A new blog post!, points to AlphaZero paper, and Lc0 v0.19.1 (blog.lczero.org) • 'The paper contains additional details that were missing in the original preprint from one year before. There were some aspects that were implemented in Leela differently from AlphaZero, and I'm sure we'll find some more.'

The most detailed discussion was on neutral ground (talkchess.com) with participants from the Stockfish & Leela communities, from DeepMind, and from other (mostly) knowledgeable experts:-

• 2018-12-06: Alphazero news

As for specific talking points, there were many -- too many to cover in a single blog post.