In my experience, LLMs have a hard time processing large amounts of raw data and finding patterns (even simple univariate time series) so I would be surprised if a GPT could find _real_, non-hallucinatory patterns in raw data.