Board Game Arena

Hi everyone,

I've started a youtube channel where I analyse Splendor replays using a alpha-zero style engine that has been trained to a high-level

Here is an example of the current generation

https://youtu.be/d4lpbIGd_SA

I'm trying to continually improve the quality so any feedback is appreciated.

Also if you are interested in having your games analysed you can post them below or message me directly. Just note currently I only analyse 2-player games (no expansions) and can't analyse games with moves containing reserving from the hidden deck.

Respectfully, the narration is extremely bad.

"Player 1 does x, player 2 does y" is not interesting. I can see what moves are being played. I want to know what those moves accomplish both offensively and defensively, what future lines they are threatening, what alternatives could have been considered. When the narration does mention something like this, it's often totally random and meaningless. For example, at 5:39 the narration mentions that the gold gained from reserving a card, will make a tier 1 white card affordable. So what? Look at the state of the game -- does it matter at all that the tier 1 white card is now affordable? The very next move" player two takes [whatever]. The engine disagrees, and prefers [something else]" Why? Take a moment to explore the consequences of both moves to explain why the engine prefers something else, otherwise what's the point of even mentioning it? The AI is just saying things for the sake of saying them -- which is all AI can really do, tbh.

Check out Gothamchess youtube channel for an example of game commentary that makes good use of engine analysis while also providing useful strategic insight. You will need to put more human thought into your narration, instead of leaning so heavily on AI.

Thanks for replying.

I agree that more narration needs to be done on strategic decisions, this happens more frequently from the middle to the end of the game - I'm wondering if you watched that far or got put off before getting through. Early on there are many similarly weighted moves so its difficult to focus on one plan early on. Splendor is not as tactical as chess so it doesn't lend itself to exploring many different tactical lines beyond a depth of 1 or 2.

In terms of improvements would it be useful to not mention moves at all, and just jump quickly through a number of moves until an 'interesting point' occurs. Perhaps not even show parts of the game at all and then just jump to particular positions which can then have more lengthy strategic commentary on? I want to keep the videos under 15 mins.

You mention Gothamchess, my inspiration actually comes from agadmator who I think does this type of analysis using an engine perfectly.