Spatial Awareness Capability

Venice ai’s spatial awareness is very very bad. if you ask it to generate a story involving multiple characters it constantly has multiple characters doing things at the same time that are physically impossible. most ai’s are not great at this but venice seems to be quite bad.

I would suggest that if the model had a spacial awareness matrix that allowed it to build a stickman type (ie extremely low res) 3d model  (essentially a mental 3d map) of where people, things and other things critical to spacial constrants (e.g aproximate distances between critical points, directon of observation etc) are, with notes at important node points. the ai would be vastly improved. the ai would need to constantly update the mental 3d model as user input and its own output changes what is going on.

LLM’s are in many respects language pattern recognition systems, understanding spacial awareness is anathema to them (pretty difficult to train something focused almost entirely on language to understand space). a seperate integrated 3d mental maping component would make the ai, infinately more usefull and reliable (if it dosn’t work in the mental mapping then it won’t output BS scenario’s).

To my knowledge pretty much no major general ai has spatial awareness mental mapping built into it and it should be critical to resolving a lot of hallucination scenarios.

Please authenticate to join the conversation.

Upvoters
Status

Completed

Board
💡

Feature Requests

Date

4 months ago

Author

An Anonymous User

Subscribe to post

Get notified by email when there are changes.