The "content" param should be flexible: str or List

[BeibinLi]

For instance, the multimodal agents use a list as "content" in messages, but standard GPT-4 and Llama based models are using string.

This basic discrepancy causes issues while two different agents are communicating with each other. An interesting bug is raised by @joshkyh while creating a GroupChat environment with two LMM agents.

Tasks

Preview Give feedback

1. Update ConversableAgent and all other agents.
2. Update system_message to be compatible with both str and list
3. Update content to be compatible with both str and list
4. Update _oai_messages to be compatible with both str and list
5. Check extract_code, function, and human_input_mode accordinly
6. Update test cases accordingly.
7. Update notebooks accordingly.
8. update OpenAIWrapper.create()
9. check GptAssistant
10. test Conversable Agent with List content type

[BeibinLi]

For reference, here is OpenAI's definition of "messages": https://platform.openai.com/docs/api-reference/messages/object