Yeah like the other say it depends on the harness. "into chatgpt" could mean many things. It's likely not predicting the next tokens to emit but rather thinking "okay the data is in format x i need to map it to y, and also this is a good fit for statistical model z" and then writing and executing a python script that runs whatever statistical model on the inputs and then produces some other output file that it then uses to produce a report.
That being said, the correct way would be to just use the LLM to produce the script, validate it themselves (often times there's something very minor wrong), and then use that script next time instead of an LLM to rewrite the same script over and over every time you need to produce a report.
That being said, the correct way would be to just use the LLM to produce the script, validate it themselves (often times there's something very minor wrong), and then use that script next time instead of an LLM to rewrite the same script over and over every time you need to produce a report.