4 ms·
> because AlphaZero works much more similar to humans than alpha-beta search. Just curious, what do you base that on? Is there a relevant article? If so I'd
by radarsat1 4y ago
> because AlphaZero works much more similar to humans than alpha-beta search.
Just curious, what do you base that on? Is there a relevant article? If so I'd love to read it.
- xiphias2 4y agoI think the best is just to listen to videos of Demis Hassabis (Lex Friedman interview was great as well). He loves to talk about how much it shocked the go community when AlphaGo created new moves / strategies that humans could learn as opposed to just brute force search algorithms. And of course it translated to chess as well somewhat. https://www.fnac.pt/Game-Changer-AlphaZero-s-Groundbreaking-Chess-Strategies-and-the-Promise-of-AI-Matthew-Sadler/a6682646 https://www.fnac.pt/Game-Changer-AlphaZero-s-Groundbreaking-...
- hackinthebochs 4y agoBroadly speaking, NN based engines make more human looking moves compared to rule-based evaluations: https://www.cs.toronto.edu/~ashton/pubs/maia-kdd2020.pdf https://www.cs.toronto.edu/~ashton/pubs/maia-kdd2020.pdf
- xiphias2 4y agoIt looks cool if Maia can make difference between a human vs Leela playing, it would be interesting to see if it can find cheating.