5 ms·
This is very out of date. They now often trigger tooling and return the outputs of the tooling.
by MattGaiser 1y ago
This is very out of date. They now often trigger tooling and return the outputs of the tooling.
- blastonico 1y agoNot really. You're referring to agents, but the model doesn't always require agents, and the public chatbot is not connected to a shell freely evaluating arbitrary commands.
- blincoln 1y ago> and the public chatbot is not connected to a shell freely evaluating arbitrary commands. [ citation needed ]
- brookst 1y agoThis is false. Modern chatbots use tool orchestration where the LLM is given a list of tools it can use. Tools include things like searching the web and, yes, executing code. The LLM can generate arbitrary code (typically Python) and call a tool to execute it and return results. The tool takes the Python input, launches it (in a container in this example) and returns the output, which the LLM can further process.