They can - if we split it into parsing, execution and response. If LLMs only do parsing and response and execution is limited to actions that are prepogrammed,

They can - if we split it into parsing, execution and response. If LLMs only do parsing and response and execution is limited to actions that are prepogrammed, i think we’ll have a much better assistant. Like, it could be interactive and natural. But i agrees that it shouldn’t be given unfettered access to do whatever it wants.