Confusing error message when `llm_config` is not initialized or is `None`.

[ekzhu]

Is your feature request related to a problem? Please describe.

Right now, if an agent's llm_config is set to None (default in ConversableAgent and AssistantAgent), or if config_list is empty.

from autogen import UserProxyAgent, AssistantAgent

# assistant = AssistantAgent("assistant", llm_config={"config_list": []})
assistant = AssistantAgent("assistant")
proxy = UserProxyAgent("user")

proxy.initiate_chat(assistant)

We get confusing error message from the openai client:

TypeError: Missing required arguments; Expected either ('messages' and 'model') or ('messages', 'model' and 'stream') arguments to be given

Describe the solution you'd like

We should raise an error early if the llm_config is None or config_list is empty, as it is guaranteed to lead to error as the model is not specified.

Alternatively, we should set llm_config by default to false in ConversableAgent and make llm_config a required argument in AssistantAgent. As the current default None is always going to cause error.

Additional context

Related:

#1210

[afourney]

Yeah, this is another case where None or empty is taken as a positive (in this case that you want to use an LLM)

Code_execution_config was the same until late last week.

At one point the default was to assume gpt-4 and read the key from the OPENAI_API_KEY. So some users setting llm_config to None and getting billed for GPT-4 use... This error is at least better than that!

[ekzhu]

@gunnarku here you go. You can investigate based on the PR description and see what is a good way to fix this.