- Anna Bernstein is a fast-paced engineer at Copy.ai, which creates AI tools to generate messages and emails.
- Its job is to write prompts to train the bot to generate accurate, high-quality writing.
- Here are three tips for writing prompts to get the best results from AI.
This as-told-to essay is based on conversations with Anna Berstein, a 29-year-old prompt engineer at New York-based generative AI company Copy.ai. The following has been edited for length and clarity.
When I was a freelance writer and historical research assistant, I spent a lot of time scrolling through microfiche in libraries. Now I am a fast engineer helping to optimize the most advanced technology in the world.
My journey into rapid engineering began in the summer of 2021, when I met a man in a jazz bar who was working for Copy.ai at the time, which creates an AI tool that can generate text for blogs, sales emails, and social media. media messages.
He said that Copy.ai – running on OpenAI’s GPT-3 language model – was having issues with output quality and asked if I’d like to try being a punctual person. I didn’t like the stress of freelancing – besides, it seemed fascinating – so I said yes, even though I studied English and had no technical background.
Soon after, I was offered a one-month contract to work on performing different types of tones. At first I barely knew what I was doing. But then the founder explained that prompting is like writing a spell: if you cast the spell just a little bit wrong, something bad can happen – and vice versa. Based on his advice, I managed to come up with a solution for better tonal stability, which led to a full-time job at the company.
Since then the scope of my work has grown. I now help improve existing tools and create new ones with the goal of getting the AI to spit out the best answers for users.
In practice, I spend my days writing text-based prompts, which I can’t disclose due to my non-disclosure agreement, which I feed into the back-end of the AI tools so they can do things like generate a high quality blog post, grammatically correct and factually correct.
I do this by designing the text around a user’s request. In very simplified terms, a user types something like, “Write a product description about a pair of sneakers,” which I receive on the back. So my job is to write prompts that can make sure that query generates the best output by:
- Instruction, or “Write a product description about this.”
- Example following, or “Here are some good product descriptions, write one about it.”
Besides the pure prompt engineering part of my job, I also advise on how the models behave, why they behave the way they behave, which model to use, if we can make a specific tool and what approach to take. to do that.
I like the “mad scientist” part of the job where I can come up with a stupid idea for a prompt and see it actually work. As a poet, the role also feeds my obsessive nature with impromptu language. It’s a very strange intersection of my literary background and analytical thinking.
However, the track is unpredictable. New language models are coming out all the time, which means I always have to adjust my prompts. The work itself can be tedious. There are days when I obsessively change and test a single prompt for hours – sometimes even weeks at a time – just so I can get them to work.
At the same time, it’s exciting not knowing what’s to come.
Aside from people at parties who don’t understand my work, one of the big misconceptions I’ve noticed about AI is the idea that it’s sentient when it’s not. When it tries to talk about being an AI, we freak out because we see so many of our fears reflected in what it says. But that’s because it’s trained on our fears, based on creepy AI sci-fi images.
While writing good prompts is easy to pick up, it’s hard to master. Getting the AI to do what you want takes trial and error, and over time I’ve picked up weird strategies; some of my prompts are really wild in texture.
Here are some tips that can help you develop better prompts:
1. Use a thesaurus
Don’t give up on a concept just because your initial formulation didn’t produce the desired result. Often the right word or wording can unlock what you do.
2. Watch your verbs
If you want the AI to fully understand your request, make sure your prompt contains a verb that clearly expresses your intent. For example, “Rewrite this so it’s shorter” is more powerful than “Condense this.”
3. ChatGPT is great in design, so use that
Introduce clearly what you are trying to do from the start and play around with wording, tenses and approximations. You can try “Today we’re going to write an XYZ” or “We’re trying to write an XYZ and we’d like your input.” It’s always helpful to put an umbrella of intent over what you’re doing, and playing around with different ways of doing that can make a big difference.