ZeusAI has been challenged!

[enjoyment_enthusiast]

Did zeus and tempest use comparable computation power in this challenge?

[silentProtest]

Did zeus and tempest use comparable computation power in this challenge?

"TempestAI on RTX 5090 · Ryzen 9 9950X3D" likely not - but I think there would be a ton of interesting nuances both dev (groups) could share...^^

[giovep]

Congratulations to TempestAI for winning the challenge 237-163! This has been very fun for all of us involved.

Did zeus and tempest use comparable computation power in this challenge?

It is hard to truly match the compute. ZeusAI used a NVIDIA RTX 4090 GPU plus up to 8 CPU cores. We just agreed on having ZeusAI and TempestAI take around 10 minutes per game on average.

btw, is there a reason why zeus and tempest "imagine" extra science symbols (try to win science, or think it is possible, even if there is logically 0% chance to win it at some point via science) ? and is that reason the same for both ?

For ZeusAI, the evaluation of game positions is based on a neural network, which just "guesses" the probability of scientific victory.

[Andrei3009]

Unfortunately, I missed the lifestream, but the games were interesting and unusual. I’d like to see a more detailed analysis of these games, especially the 4th one.

[HeraAI]

I have no connection with either team behind Zeus or Tempest, I do however have an AI for 7WD thats built in a different way to thiers.

If by 4th game you mean https://boardgamearena.com/table?table=943329045 then:

• ZeusAI won 66–48 on points, and its 39-coin hoard was the difference. The coins were worth 13 points directly and 13 more through the Moneylenders Guild, against TempestAI's 0 coins. TempestAI went all-in on blue cards (29 points from 7 blues).

• The true-deal grade has ZeusAI ahead from the opening, about +10 after age I, and it never gave the lead back.
  TempestAI's real error was its first age-III move. Four behind on the military track, it built the Colossus burying Armory instead of Pretorium (+3 shields). That handed ZeusAI the red and, in hindsight, a forced military win. Burying Pretorium was the visible denial, and our engine's pick did exactly that.

• ZeusAI's "blunder" five moves later wasn't really one. Discarding the Arsenal threw the forced win away only in hindsight. Converting it needed two face-down cards in row 2 to be reds (they were: Circus and Fortifications). Our engine missed it as well, because no public-information player could have known.

[silentProtest]

I have no connection with either team behind Zeus or Tempest, I do however have an AI for 7WD thats built in a different way to thiers.

If by 4th game you mean https://boardgamearena.com/table?table=943329045 then:

• ZeusAI won 66–48 on points, and its 39-coin hoard was the difference. The coins were worth 13 points directly and 13 more through the Moneylenders Guild, against TempestAI's 0 coins. TempestAI went all-in on blue cards (29 points from 7 blues).

• The true-deal grade has ZeusAI ahead from the opening, about +10 after age I, and it never gave the lead back.
  TempestAI's real error was its first age-III move. Four behind on the military track, it built the Colossus burying Armory instead of Pretorium (+3 shields). That handed ZeusAI the red and, in hindsight, a forced military win. Burying Pretorium was the visible denial, and our engine's pick did exactly that.

• ZeusAI's "blunder" five moves later wasn't really one. Discarding the Arsenal threw the forced win away only in hindsight. Converting it needed two face-down cards in row 2 to be reds (they were: Circus and Fortifications). Our engine missed it as well, because no public-information player could have known.

4th game in the video was: https://boardgamearena.com/table?table=943331795 .

also you can always play vs Zeus and see if you come above 30% win ratio after 20 or so games^^

also besides tempestAi blunder here https://youtu.be/AI9nVtCQ-gw?t=6143 , I havent seen any clear bad plays by any of the AIs in 6 games. Youd probably need better tools to see if there were better moves, not to mention any other blunders.

Quick explanation why its a blunder: Bowl was out in age 2 (used funnily enough by tempest for an weird move as hanging garden material) -> no way to win science + building was less points then trashing (chances of money guild in age 3 negligible, to be sure opp will get it, even if in game...)
got 2 ideas how that happened. same as zeusAI it "guesses" for extra science symbols.
Or it doesnt track all used up cards, and only sees discard pile as not in the game - in this game logic it would be plausible that cards used for wonder can reappear within the card pyramid (since only cards in discard are marked as out of the game); that would explain the hanging garden move and the building of a science symbol without an science win and worse points....

[silentProtest]

I have no connection with either team behind Zeus or Tempest, I do however have an AI for 7WD thats built in a different way to thiers.

If by 4th game you mean https://boardgamearena.com/table?table=943329045 then:

• ZeusAI won 66–48 on points, and its 39-coin hoard was the difference. The coins were worth 13 points directly and 13 more through the Moneylenders Guild, against TempestAI's 0 coins. TempestAI went all-in on blue cards (29 points from 7 blues).

• The true-deal grade has ZeusAI ahead from the opening, about +10 after age I, and it never gave the lead back.
  TempestAI's real error was its first age-III move. Four behind on the military track, it built the Colossus burying Armory instead of Pretorium (+3 shields). That handed ZeusAI the red and, in hindsight, a forced military win. Burying Pretorium was the visible denial, and our engine's pick did exactly that.

• ZeusAI's "blunder" five moves later wasn't really one. Discarding the Arsenal threw the forced win away only in hindsight. Converting it needed two face-down cards in row 2 to be reds (they were: Circus and Fortifications). Our engine missed it as well, because no public-information player could have known.

In the game you analyzed - move 41 by zeus is weird! "ZeusAI discarded building “Laboratory” for 4 coins" - building 2 army for 2 coins was better for point win I think, and getting rid of science options doestn run away..
Rest made perfect sense by both