The Art and Science of AI Prompt Engineering
The key prompt engineering skill for generative AI may be understanding how different models pay attention to different parts of a prompt rather than crafting the "perfect" prompt.
Summary
- Prompt engineering has emerged as an important skill for getting good results from generative AI tools.
- However, different language models pay attention to different parts of a prompt - some focus on the beginning, some the end, some the middle.
- So crafting an ideal prompt requires understanding precisely how each model handles prompts.
- This makes prompt engineering much more complex than with past technologies like Boolean search.
- We may have to wait for more scientific testing to understand prompt interactions before we can effectively train lawyers on prompt engineering.
- The legal industry's oscillation between calling the technology "AI" and "machine learning" mirrors what happened more broadly in tech.
- Issues like summarization could fail if models ignore parts of documents that lawyers might expect to be incorporated.