Guidance on drafting expert prompts on the Screens platform, with a focus on creating clear, effective instructions for large language models (LLMs).
1. Purpose of Expert Prompts
The primary goal of drafting expert prompts on the Screens platform is to leverage the capabilities of today's large language models for solving contract review problems. The effectiveness of these prompts lies in the ability to provide precise and unambiguous instructions to the LLM.
2. Components of a Prompt
2.1 Standard or Question Text
The standard or question text is the visible part of the prompt, residing on top of the platform. It is the initial set of instructions that users see and should be crafted to be clear and understandable.
2.2 AI Guidance
The AI guidance, although more under the hood, is also accessible to users. It allows fine-tuning of instructions for the LLM, enabling the creation of defined terms and exceptions to enhance the model's understanding.
3. Framework for Expert Prompts
3.1 Viewing LLM as a High-Powered Associate
Consider the LLM as a high-powered associate with legal reasoning capabilities but lacking contextual understanding. Avoid broad and vague instructions that require extensive context.
Example of broad or vague instruction:
Make sure that the vendor's liability falls within our usual parameters in this vendor contract.
3.2 Task Breakdown
Instead, break down tasks into small, clear, and concise chunks without ambiguities. This approach facilitates better comprehension by the LLM, leading to more accurate and relevant results.
Example:
- The vendor's liability should be capped at ≥ 12 months fees
- Damages or claims relating to confidential information and the customer's IP should be excluded from the cap
- The vendor should indemnify the customer for third party IP infringement
- The vendor's indemnification obligations should be excluded from the ....
4. The Four C's
A crucial framework for drafting effective prompts is the adherence to the four C's: Clear, Context, Concise, and Small Chunks.
4.1 Clear
- Objective: Avoid vagueness and ambiguity.
- Application: Ensure that instructions are explicit and easily understood by the LLM and users.
4.2 Context
- Objective: Provide explicit context for instructions.
- Application: Avoid assumptions about the LLM's understanding; use defined terms to add clarity when necessary.
4.3 Concise
- Objective: Reduce noise in instructions.
- Application: Minimize unnecessary information to facilitate easier comprehension by the LLM.
4.4 Small Chunks
- Objective: Break tasks into bite-sized pieces.
- Application: Divide complex tasks into smaller, manageable chunks to enhance the LLM's ability to process and respond accurately.
5. Conclusion
By following the principles outlined in this document, users can become expert prompt engineers on the Screens platform, transforming their prompts into powerful tools for achieving precise and reliable results in contract review tasks. Adhering to the four C's—Clear, Context, Concise, and Small—will contribute to optimal performance on the platform. Users are encouraged to refer to these guidelines while drafting prompts and to revisit them when analyzing prompt performance.
Best of luck in your prompt engineering endeavors on Screens!