BlighterDalekBots~in World Variant games?

[MajorMitchell]

Are there any plans to expand the opportunities to have FUN playing against hordes of BlighterDalekBots in the World variant game at Webdip?

Playing against six Naughty BlighterDalekBots in Classic is great fun but six of those algorithmic fiends are relatively easy to defeat.

One human player against sixteen BlighterDalekBots in World games will prepare young whippersnappers for the Bwave New World of AI in which they will live most of their lives.

It will also give Ageing Veterans who started life playing with Crystal Radios & then
had to survive the Western Front & innumerable conflicts since those halcyon days of glorious youth the opportunity to demonstrate the superiority of the human intellect over it's creations.

[MajorMitchell]

Bumpity bump.

[kestasjk]

Unfortunately that's pretty unlikely to happen MM: We can tweak the parameters of the AI model, e.g. make them easier/harder, chattier/quieter, less/more rude, truthful/deceptive, etc, but we can't (without buying supercomputer time and a *lot* of effort) train new models that would know about a different map.

And even if we had the know-how and hardware to train new models you still need a huge dataset to train with: the Classic map naturally has by far the largest dataset, and I would expect World would have far fewer games, and have more parameters to train on due to more territories and countries.

So I'm afraid I don't think it'll happen, only for the most frequently played variants, with data from vDip, and with huge effort.

[wulfheart]

Unfortunately that's pretty unlikely to happen MM: We can tweak the parameters of the AI model, e.g. make them easier/harder, chattier/quieter, less/more rude, truthful/deceptive, etc, but we can't (without buying supercomputer time and a *lot* of effort) train new models that would know about a different map.

And even if we had the know-how and hardware to train new models you still need a huge dataset to train with: the Classic map naturally has by far the largest dataset, and I would expect World would have far fewer games, and have more parameters to train on due to more territories and countries.

So I'm afraid I don't think it'll happen, only for the most frequently played variants, with data from vDip, and with huge effort.

I think it might work by letting them play against each other. But chat messages would be difficult. However, I believe there is a space for research on transforming the output of e.g. DAIDE orders/structured communication into text. If I had more time it would be so interesting. I would also have the ability to use the GPU cluster at my university. It is quite new and at least some months was mostly idling around because of reasons I don't comprehend.