12
Ran GPT-4 against Claude 3.5 for my HVAC diagnostic logs, the winner was obvious
I do a lot of side work troubleshooting commercial RTUs and I've been feeding service call notes into both models to see which one catches refrigerant issues better. GPT-4 kept giving me generic "check the filter" suggestions even when I told it superheat readings. Claude caught a bad TXV valve from my description of a 12 degree superheat on a 95 degree day in Phoenix. Not even close. Has anyone else noticed one model clearly beating the other for specific niche work?
3 comments
Log in to join the discussion
Log In3 Comments
miles_burns1mo ago
Saw a tech on YouTube break this down with refrigeration data and Claude picked up on liquid line restrictions way better than GPT. Crazy how much the small details matter for stuff like TXV failures.
3
sage_green1mo ago
I mean, the small details with TXV failures are exactly why I'd trust a Claude over my own brain at this point. Last time I tried diagnosing a liquid line restriction, I spent 45 minutes chasing a ghost while the unit was just low on refrigerant. Maybe it's just me but I feel like every HVAC problem has three possible causes and I always pick the wrong one first.
3