How LLMs work

To get the best use out of these tools, it helps to know how they work. An LLM is trained on vast amounts of available data, and coded to respond to prompts based on that training data. Each time you prompt an LLM, it runs a calculation and “rolls the dice”, referring to its training data, then comes up with a response based on the context you give it. This means that when you prompt an LLM, asking the same question may get you a different answer each time, and sometimes, the answer may be irrelevant or incorrect.

Thoughtful prompting can get you better responses. Here are some tips for prompting LLMs for better results, using the CRAFT model:

• Context: Provide background information and audience
• Role: Specify the persona or point of view
• Action: State the output or task
• Format: Specify the format
• Tone: Define the tone or style

Adding any of these in your prompt makes it more likely that you will get relevant, helpful responses. However, it’s still important to evaluate the responses that these tools provide, to verify their accuracy. You can do this in a few different ways:

• Doing a non-AI search (using platforms like Wikipedia or the Libraries website) to double-check factual responses
• If the response cites a source, go to that source and verify that the AI response correctly represents the information there