Prompting #
Increasing Accuracy #
To improve system prompts used in agents, we can apply various prompt engineering techniques.
Here are some tips to increase response accuracy of LLMs:
Personas Prompting (you are an expert) #
I often see prompts that start with “you are an expert in [topic]… please give an answer to [question]”. Apparently, this is called the “personas prompting” approach that can guide the model to answer more accurately.
Few-Shot Prompting (Example-Driven) #
We can provide examples of the input and some good-quality output. This can probably guide the LLM toward the patterns in its training space where a similar quality data is present (that’s my guess how it can improve the generation).