8
Tried OpenAI's o1 for coding and it flopped compared to Claude 3.5 Sonnet on a real project
Honestly, I gave OpenAI's new o1 model a shot for a Python script I was writing to scrape and clean a bunch of messy CSV files last Tuesday. Everyone was raving about its reasoning abilities, but it kept hallucinating column names and suggested a convoluted regex approach that broke on the second file. Then I switched to Claude 3.5 Sonnet, and it got the whole thing right in one pass-clean pandas code, no errors, and it even caught a data type mismatch I missed. The difference was huge: Claude just understood my messy real-world data better, while o1 felt like it was showing off textbook solutions that didn't fit. Has anyone else found Claude beats OpenAI's newer models for actual hands-on data work?
2 comments
Log in to join the discussion
Log In2 Comments
miam111mo ago
Three hours on a fake column? Sounds like he needs better backup habits.
2
harperwright1mo ago
Heard my buddy Dave spent three hours debugging an o1 suggestion that added fake columns to his sales data, then Claude fixed it in ten minutes. He was so mad he almost threw his laptop out the window.
0