How to use prompt engineering

Here is a quick look at each:

• AI stands for artificial intelligence or augmented intelligence, and it's a field of computer science that creates systems capable of performing tasks that typically require human intelligence. Examples include facial recognition software, voice recognition systems, customer service bots, and recommendation algorithms
• Generative AI, or GenAI, is a subset of AI that teaches machines to produce content, including text, images, music, or video. Examples include OpenAI's ChatGPT, DALL-E image generation, Microsoft's Co-pilot, Thomson Reuters' CoCounsel, Google Gemini, and many others
• Large Language Models are a specific type of GenAI focused on understanding and generating human language. Trained on vast amounts of data, LLMs can do several tasks, including being used in GenAI to respond to prompts. By analyzing trillions of sentences, LLMs can learn enough to finish an incomplete sentence, or even generate a sentence based on context

Clarity, specificity, and context. You want Generative AI to understand the task. If you are asking about a legal scenario, include the jurisdiction, area of law, and pertinent case details or legal principles.

There are, and your results will vary from tool to tool, depending on the training database and the algorithms used. The tone of responses will also vary. You can experiment by asking the same question of different systems. If you are doing legal tasks, we recommend using a specialized AI designed for legal tasks and always reviewing the terms and conditions of whatever product you choose.

Tell the GenAI model exactly what you're looking for by being clear, specific, and providing the correct context. For example, specify that you want output as a bullet-point list, a summary of a desired length, a detailed analysis, code in the R programming language, or structured data.

Very possibly. If you are not paying for the product, then you are the product, and what you upload can be used for training the system. If you are paying for the tool, read the terms carefully to understand how your information may be used.

Asking a general knowledge question carries little or no risk. However, information that is confidential or proprietary carries a high risk and should never be entered into a GenAI model that does not have strong safeguards. If you use our AI sandbox, what you enter is not being used for training the system but is visible to our staff.

Yes, and that's why all AI output should be verified. If a GenAI tool does not have access to the relevant data or if you have not given a specific, clear prompt with the proper context, it may guess what you want. AI answers can still sound authoritative even if they're incorrect. Examples include citing case law that does not exist, and this kind of false information is referred to as "a hallucination."

They do, with one known issue being regurgitation, which is the tendency of Large Language Models to provide responses that are identical or very similar to the language the model was trained on. This could result in unintentional plagiarism or copyright infringement, so it's important that your prompt indicates you want to avoid regurgitated responses.

Here are some strategies for better prompts:

• Try using a back-and-forth approach, also known as iterative refinement. To do this, start with a basic prompt and then ask follow-up questions or provide instructions to delve deeper or correct misunderstandings.
• Be aware that your AI tool has a "context window," and once you reach the limit of the context window, the AI will "forget" what it has been told previously in the chat. If you encounter this issue, you can remedy it by including the relevant details from earlier in the chat in your next prompt.
• You can also ask the AI how to craft an effective prompt for the task. Known as "meta-prompting," this technique can be powerful because it leverages the AI's knowledge of concepts and terminology.
• Take time to experiment with different phrasings and approaches. What works well for one task may not be ideal for another, so it's worth it to develop prompt templates to help handle your common workflows.

Here are a few examples of good and bad prompts:

• For a motion to dismiss:
  • Good prompt: Draft a motion to dismiss for lack of personal jurisdiction in [jurisdiction] based on the following facts: [list key facts]. Include relevant case law citations and avoid regurgitated responses
  • Poor prompt: Write a motion to dismiss
• For a contract review
  • Good prompt: Compare and contrast the elements required for a valid contract in [State A] versus [State B]. Organize the response in a table format
  • Poor prompt: What are contract laws?
• Understanding a civil case
  • Good prompt:  For the case of [insert name of case], summarize the main arguments from both sides, focusing on contract interpretation, and provide a list of key precedents
  • Poor prompt: Tell me about the case
• Writing a job description
  • Good prompt: Please write a job description for a Trial Court Administrator. After the description, please provide evaluation criteria for that position
  • Poor prompt: Write a job description