Jev gets some information about itself (xyz, hunger, dimension), inventory, nearby blocks, nearby entities, tell it if it's about to walk into anything in any cardinal direction, and some memories.

Homeby PLx222026-09-16

Jev gets some information about itself (xyz, hunger, dimension), inventory, nearby blocks, nearby entities, tell it if it's about to walk into anything in any cardinal direction, and some memories. For now, JevBot also gets 6 20x20 ascii text representations of what would be seen North/south/east/west/up from its position... the 6 ascii windows trick doesn't work *great* though. Definitely curious to see what works for perception. I found giving it a bunch of "common sense" context about how to play minecraft is really helpful. Like "Solid blocks block movement" and "Mining/harvesting may require a suitable tool". After this, JevBot gets a Choice over a list of legal actions from the current state. These are things like "collect snow", "fight witch", "craft stick", etc.

open source

More in Home

I tried to create a latent context layer that is a warmed up version of a small LDA Topic model that is injectable into the context window via memories or system prompts.I made some CVSS classification scripts that use Jev to generate a CVSS vector for a given software vulnerability description.Jev can be used for profiling players, which is great for organizing playtests and checking balance changes across different player groups.