Comment by leevilux

Wouldn't self-improvement mean that the LLM changes its neural network (i.e. the weights or layers or back propagation algorithm etc) or modify its training data?