Rachel Fernandez, Director of Sales Engineering
With the rise of generative AI and the proliferation of various tools based on large language models (LLMs), a field referred to as prompt engineering blossomed. Individuals with expertise in prompt drafting for LLMs were an immediate need. As technology has become more readily available, the need for a subject matter expert has waned. Instead, the development of technology has focused on lowering the barrier to entry when using these tools. While accessibility is improving, the basic tenants of prompt engineering can still positively impact your AI outputs.
At Sandline we follow a CLEAR method when writing prompts. The CLEAR mnemonic where each letter of the word CLEAR helps you to remember an important keyword and association for prompt generation.
- Context: Always provide clear background information. Essential details, such as legal claims or details that drive relevance can greatly enhance your prompt output. Describing field specific acronyms, key individuals and their roles, and important events will greatly improve the LLMs’ ability to understand the overall matter.
- Language: Use plain, natural language. LLMs often struggle with dense legal jargon unless it is clearly explained. Define the details, acronyms, and technical terms the same way you would describe it to a summer intern.
- Examples: Provide the models with examples of what you are hoping to find in the data set. Provide ideal outcome examples to help it learn what you want.
- Affirmative: Models respond best to affirmative language. Negative phrasing and double negatives can confuse your output. Avoid “don’ts” in your instructions. For example, it is better to state that non-relevant documents contain party invitations instead of relevant documents that don’t contain party invitations.
- Refine & Repeat: Prompt writing is iterative. It is a process of trial and error. Expect to do it more than once.
Learn what your model isn’t understanding and adjust your prompt accordingly.
To further improve your prompts, it’s important to learn about the LLM powering your tools. Some are more sensitive to formatting than others. Each LLM may have its preferences for certain phrasing or specific details. The more you learn about your LLM, the more effectively you can teach it. This CLEAR method is an agnostic tool and serves as a baseline for prompting your generative AI tools.