The attempts I have seen to use large language models with parser IF, such as LampGPT, seem to be all about improvising new responses to player input, rather than going the more obvious route of only intervening when the standard parser fails to understand the input, translating it into something the parser understands.
1 Like