cybergod-ideas-and-innovations.txt

 
i have many ideas which require implementation and verification, like adding a token classifier before embedding to token output/resolution (is this a thought token, imagination token, or action token), simutaneous token generation for actions, imagination (reality projection) and thoughts, periodic multi-layer multi-size multi-duration temporal memory update and cleanup to prevent context overflow for long time tasks, latent memory space, real time non-blocking bidirectional dialogue system, real time learning (model parameter update) etc. i am creating the verification parts. currently with ctf-gym, vimgolf-gym. the cybergod-gym is for checking if the agent can earn money at a constant speed. now the mulerun is almost capable for that. so what i can say is to observe for active developments, adapt others training/inference frameworks, benchmarks and models, innovate and verify.