What is Prompt Engineering?
Prompt Engineering is the process of discovering the right words to use so that an AI generates the result you want. It’s being used a lot these days with image generation AIs like DALL·E and Mid journey.
Waverly also uses Prompt Engineering, but in our case, the goal is to help an AI generate the content feed you want. You can write a few sentences describing a topic you’re interested in — with as much detail as you want — and our AI will strive to understand it. From there, its goal is to provide a well-curated content feed as the output. A feed without noise, without distractions.
Naturally, AI being what it is, Waverly might not understand every subtle detail of what you wrote. This is where you might want to put your engineer hat on and try variations, repeat a word that is important, cite a few sources you care about, etc. We’ll get into these details later.
But before we dive into how you can benefit from prompt engineering, it is essential to understand how Waverly works.
How does Waverly work?
There are two main components in Waverly’s technology. The first one is the Waverly AI document model. Our model sifts through millions of blog posts, news articles, and industry reports to surface the information you need. It analyses each of these documents with the latest natural language understanding technology to extract what the article is really talking about.
The second component is the wave, a structure you provide to the Waverly AI so it can continuously provide intriguing and relevant content. This wave contains a title, a summary and a list of text prompts. Together, they allow the AI to understand your topic of interest in as much detail as you want.
The title and summary are limited in size. Their goal is to give a brief overview of your topic of interest so anyone stumbling on your wave (if it’s public) can learn what you’re interested in.
The prompts have no such limitations and can be as long as you want. Their goal is to help the AI get a better understanding of your needs. They give you the power to steer the recommendations in the direction you want. Prompts can be very long or very short. However, if you have a specific outcome in mind, you will need to be specific and this is where prompt engineering enters the equation.
Prompt Engineering — how to talk to our AI?
Like any process, the quality of the input determines the quality of the output. Figuring out how to write good prompts can significantly improve the quality of the article feed you’ll get from the Waverly AI.
Tips to engineer the best prompts
- Spelling errors and grammar sometimes confuse our AI. Get your text through a spellchecker to help our AI read your text accurately.
- If you want to emphasize a keyword in your prompts, repeat the keyword a few times. You can even provide the full definition of the concept — in your own words, or by cut-and-pasting a paragraph from Wikipedia.
- Continuously tweak your prompts based on the content you get. If you feel your feed doesn’t contain enough articles on a subtopic, add a prompt that describes this subtopic.
- Try some variations Writing the best prompt involves trial and error. “Investing outside of the U.S,” “investing in developing countries,” vs “investing in foreign markets” may sound similar, but our AI understands them slightly differently and will adjust its results based on the version you use. You can even add all three of them!
- Focus on the positive. At the moment our AI doesn’t understand negative sentences. So instead of saying what you don’t want in your feed, spend more sentences describing what you want more of.
Other good practices
- Tap the fit and unfit buttons to tell our AI whether it is doing a good job or not. Our AI is continuously exploring around your topic and it will help it better understand what you want.
- Remove sources you don’t want to see on your wave. To do this tap the ⋮ button when an a article from this source reaches your wave and select Hide source from this Wave.
Sign up to join our mailing list at mywaverly.com and download Waverly on your iPhone. If you are a business interested in deploying Waverly in your organization, contact us at email@example.com