Interested in knowing specific use cases for e-commerce search:
- How to use LLM output to show more relevant products?
- How to use prompts for retrieval and ranking from existing catalogs.
- How to ensure consistency for LLM output.
1 Like
@nikhil.kulkarni,
In #1, if you’re asking how to get an instruction-tuned LLM, like chatGPT (either through the UI, or programmatically, as in this course) to give you more relevant results, then you’ll want to use the first principle explained in the Guidelines lecture - use clear and specific instructions. If, after your first attempt, you find your results aren’t as relevant as you want, then you’ll need to iterate and try again with a more specific request. The idea of iteration is discussed in the “Iterative” lecture.
For #3, one thing that can help with consistency that you’ll see in the “Expanding” lecture is the concept of “temperature” - by setting the temperature to 0, you’re saying you want a more consistent result, but with a high temperature (closer to 1, like 0.7, you’ll get more variety introduced into your result. With a temperature of 0, it will always choose the most common next word, but with higher temperatures, it might occasionally choose the second or third, etc. most likely next word.
The course aims to give you some general concepts and get you thinking about ways you can use the model. As you go through the lectures, take time to try out the examples, and experiment, as well as thinking about how you might apply these concepts to real-world problems you’d like to address.
Happy learning!