Ran a side-by-side test of GPT-4 against Claude for drafting emails

I gave both the same rough notes from a client meeting last Tuesday. Claude got the tone right on the first try, while GPT-4 needed two rewrites to sound less robotic. Has anyone else noticed a big difference in writing style between these models?

2 comments

2 Comments

jana_hill2727d ago

Honestly it's wild because Claude just feels like it actually understands how humans talk to each other, while GPT still comes off like it's trying to impress a tech demo crowd. Makes me wonder if real-world conversation data matters way more than fancy benchmarks.

cameron_palmer27d agoMost Upvoted

In my experience the difference really shows when you test them on stuff that feels like it came from a group chat rather than a scripted Q&A. I remember trying to get GPT to explain why my friend was mad about something totally trivial and it overanalyzed everything like it was grading a test. Claude just got the subtext like someone who actually lives with humans. Your mileage may vary but real conversation data seems to matter way more than people give it credit for.