Generate Knowledge Prompting
LLMs continue to be improves and one popular technique is to integrate knowledge or information to help the model make more accurate predictions.
Using a similar idea, it was tested if the LLM can generate knowledge before making a prediction in this Research Paper
Let’s test it
Prompt:
Part of golf is trying to get a higher point total than others. Yes or No?
Answer
Yes.This example demonstrates the limitations of the LLM to perform tasks that require more knowledge about this world. In Knowledge Generation with given no previous knowledge the model will give responses which almost feel correct but are absolutely wrong.
Upon giving it more information or integrating knowledge the LLM will be able to answer in a better way! Although confidence in the answer can vary depending on the amount of knowledge.
This method in Prompt Engineering can be used in Answering QA Questions by integrating knowledge into the prompt, such a use case is in SAiC (Smart AI Chat Bot)
References