4
I was sure fine-tuning a model from scratch was the only way to get good results for my project.
After comparing a heavily fine-tuned Llama 2 model against a few well-crafted prompts on GPT-4 for a specific customer service task, the prompt-based approach not only performed better but saved me over $800 in compute costs and three weeks of development time, so what's the most complex task you've seen handled well with just clever prompting?
3 comments
Log in to join the discussion
Log In3 Comments
dakota_gibson23d agoMost Upvoted
That Wired article on prompt-chaining legal docs was wild...
8
the_tyler23d ago
Man, I felt that. Honestly, reading about lawyers chaining together prompts to draft contracts gave me a real headache. It's just another thing to worry about getting wrong. Tbh the whole AI law thing is moving way too fast for comfort.
1
henrygrant5d ago
Yeah, that "moving too fast" part is exactly why I just use simple checklists now.
5