Hi,
I would like to create an agent that analyzes CSV files coming from Tableau in order to identify irregularities. After creating the agent, I noticed some strange things. When I asked it to calculate averages, I saw that it was making errors with the sum.
How is this possible?
How can I fix this, and what is the reason it didn’t work correctly?
Thank you for your help, I’m fairly new to this field.
for example :
the sum is 685
the prompt :
agent_executor.invoke({“input”: f"what is the exact averages for any list : {response.list_of_value_per_segment}, think step by step and show the calculations "})
I used ChatGPT 4o
thanks
Hi there,
This is a common misunderstanding, Large Language Models, while they run on a computer, are not “computers” they have very similar limitation with numeric processing as humans do.
You are better asking the Model to extract the numbers into a json object and then processing that with traditional code, better still would be to see if your input data follows a common format and if it does use traditional code for all of that stage to both extract and process it.
You can give models tools to use like a calculator function, but again this is only really practical for limited input datasets, large volumes of numeric data are not well suited to LLM processing at this stage in their evolution.
Thank you very much for your response.
Unfortunately, my data doesn’t follow a common format.
I added a tool for averaging lists, but you’re right, it didn’t help.
Would switching to the o1 model help?
Otherwise, if I understand correctly, only traditional code could assist me?
Thanks in advance.
