5 ms·
Only solution is to train the issue for the next time. Architecturally focusing on Episodic memory with feedback system. This training is retrieved next time
by rush86999 7mo ago
Only solution is to train the issue for the next time.
Architecturally focusing on Episodic memory with feedback system.
This training is retrieved next time when something similar happens
- atarus 7mo agoTraining is an overkill at this point imo. I have seen agents work quite well with a feedback loop, some tools and prompt optimisation. Are you doing fine-tuning on the models when you say training?
- rush86999 7mo agoNope - just use memory layer with model routing system. https://github.com/rush86999/atom/blob/main/docs/EPISODIC_MEMORY_IMPLEMENTATION.md https://github.com/rush86999/atom/blob/main/docs/EPISODIC_ME...
- atarus 7mo agoMemory is usually slow and haven't seen many voice agents atleast leverage it. Are you building in text modality or audio as well?