Hey @tutacat! My name is Tim and I work with Tyler here at Acorn Labs. Thanks for the question! This isn't quite how things work with GPTScript today, so to make it clear here's how confirm calls are generated when we start talking about how LLMs ask for permission to do something. When the LLM requests a tool call, the gptscript code will prompt the user for approval before actually executing the call. What's important to note is that all potential tool calls and their inputs are LLM generated; However, the ability for those tool calls to actually run require user approval.
Your "permissions" or confirmations should be written in code, not requiring a program to not hallucinate it, etc.
Hey @tutacat! My name is Tim and I work with Tyler here at Acorn Labs. Thanks for the question! This isn't quite how things work with GPTScript today, so to make it clear here's how confirm calls are generated when we start talking about how LLMs ask for permission to do something. When the LLM requests a tool call, the gptscript code will prompt the user for approval before actually executing the call. What's important to note is that all potential tool calls and their inputs are LLM generated; However, the ability for those tool calls to actually run require user approval.