6 ms·
Not 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 arbitr
by blastonico 1y ago
Not 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.