Tool selection doesn't help and agent is not adopting the tool for execution

[Jeffwan]

I feel it's really hard to reproduce the use cases here. https://docs.xlang.ai/use-cases/data-agent

kaggle tool is selected based on the output but not executed.. I never get lucky to get a result like what website shows (multiple kaggle dataset and I can click download to interact with)

openagents-backend-1   | 2023-11-03 22:10:53 | DEBUG - DefaultUser++65456f07e8aadbb3bc46a5d6->/chat New human message:{'message_type': 'human_message', 'message_content': 'Can you help me find datasets on Kaggle that specially focus on NBA play performance statistics over multiple seasons?', 'message_id': 37, 'parent_message_id': -1}
openagents-backend-1   | 2023-11-03 22:10:53 | DEBUG - DefaultUser++65456f07e8aadbb3bc46a5d6->/chat New ai message:{'message_type': 'ai_message', 'message_content': 'Sure, I can help you with that. Let me use the KaggleDataLoader tool to search for datasets related to NBA play performance statistics over multiple seasons.', 'message_id': 38, 'parent_message_id': 37}

openagents-backend-1   | 38.31.166.90 - - [03/Nov/2023 22:10:53] "OPTIONS /api/recommend HTTP/1.1" 200 -
openagents-backend-1   | 2023-11-03 22:10:54 | DEBUG - ++-> get_activated_message_list:[]

[Timothyxxx]

Thanks for your question! Could you try again with the GPT-4 endpoint? Just as I mentioned in https://github.com/xlang-ai/OpenAgents/issues/66#issuecomment-1793332707 , today's system with prompting mechanism rely on powerful LLMs to drive the cases

[Timothyxxx]

There are actually other ways to make it better, instead of prompting, check our tech report here.