Has anyone noticed differences between Claude 4 Sonnet and Claude 3.7?
Asked by
RandomTurtle8
Hey everyone! I've been testing the new Claude 4 Sonnet and wanted to share my thoughts. I ran into some issues while dealing with a PostgreSQL dump, and I couldn't help but notice that Claude 4 seems to hallucinate quite a bit. It suggested some nonexistent options for `pg_dump` and even incorrectly blamed Python's psycopg for my restoration problems. So, I switched back to Claude 3.7, and although it initially struggled to identify the issue, at least it didn't hallucinate. After a few tries, it managed to find the problem. Just to note, I used both models without requiring much extended reasoning. Has anyone else experienced this? It seems like things might have gotten worse with Claude 4!
Answers
I’ve been using Claude 4 since launch and I’m pretty satisfied, although I didn't do a direct comparison with 3.7. I think they’ve fixed some of the over-engineering issues in 3.7 which really helps in my case!
Reply to this answer
I think your experience might be on the extreme side! I’ve actually found the quality of Claude 4 to be better than 3.7, but I guess it can depend a lot on how you prompt it, what tools you’re using, and other factors.
Reply to this answer
Honestly, I've noticed more hallucinations too, and it sometimes acts on parts of my code that I didn't ask for, which is super frustrating. It could be a version thing. I'm tempted to go back to Claude 3.7.
Reply to this answer
Add your answer
Answers are reviewed before publication.
Related Discussions