Anyone tried automating work with Claude 3.5?
I heard Claude 3.5 Sonnet has gotten significantly better, so I'm looking to apply it to actual work. I work in the marketing team, and I'm thinking of using it mainly for automatic classification of customer response emails or drafting reports. Is there anyone using it for similar purposes? I'd appreciate any tips on prompt structure or API integration. I'm especially curious about ways to reduce hallucination.
4 answers
I'm also using Claude 3.5 Sonnet in the marketing team. The automatic classification of customer response emails is surprisingly accurate. However, the hallucination issue significantly decreased after I added the instruction 'If you're not sure, say you don't know' to the prompt. API integration was straightforward using the Python requests library.
I mainly use it for drafting reports. To reduce hallucinations, it's best to specify the output format in detail. For example, adding conditions like 'write in a Markdown table, include sources' made it much more reliable. Including 2-3 examples in the prompt also helps.
Claude 3.5 Sonnet, I'm interested in it too, and from what I've heard from actual user reviews, it seems pretty good. By the way, how did you handle labeling when classifying customer response emails? I'm also planning a similar project, so if you have any additional tips, please share them!
Oh, this is a great tool for the marketing team! I'm on the engineering team, and I tried automating code reviews with Claude, but hallucinations were sometimes an issue. However, lowering the temperature value to 0.2 or below made it much more accurate. Give it a try!