AI tips engineers on three ways to get the best results using ChatGPT

Anna Bernstein is an AI prompt engineer at Copy.ai, a company that develops AI tools that generate posts and emails.

#Bernstein’s main job is to write prompt words to train artificial intelligence robots to generate high-quality and accurately expressed articles. Here are three tips on how to write prompt words and get the best results from AI.

When Anna Bernstein was a freelance writer and historical research assistant, she spent a lot of time in the library looking at materials. Now she is an artificial intelligence prompt engineer, helping to optimize ChatGPT, the world’s largest Cutting edge technology.

Bernstein said her journey to becoming an AI prompt engineer began in the summer of 2021, when she met one of the founders of Copy.ai at a jazz bar. According to him, the company has developed an artificial intelligence tool that can generate scripts for blogs, sales emails and social media posts.

He said that the artificial intelligence running on OpenAI's GPT-3 language model encountered some problems with the output quality, and asked Bernstein if he wanted to try to become an artificial intelligence prompt engineer. Even though she was an English graduate and had no technical background, she said yes because she didn't like the financial pressure of freelancing and the position seemed interesting to her.

Soon after, Bernstein was offered a one-month contract to adjust the artificial intelligence’s different types of tones. At first, she had no idea what she was doing, and the founder later explained to her that its prompts worked a bit like writing spells: if the spell was spoken incorrectly, the AI ​​would make slight errors, and vice versa. After taking his advice, Bernstein managed to come up with a better solution for tweaking the pitch, a success that landed her the full-time job.

Since then, Bernstein has expanded the scope of his work. Now she's working on improving existing AI tools and helping the company create new ones, with the goal of making AI the best response to users.

In fact, Bernstein writes text-based prompts every day (she is not revealing the content of her prompts due to a confidentiality agreement) and feeds them into the backend of the artificial intelligence tools so that they can do Things like producing a high-quality, grammatically correct, factually accurate blog post.

Bernstein does this by designing text around the user’s request. Simply put, a user enters something like “a product description for a pair of sneakers,” and she receives it on the backend. Her job, then, is to write prompts that allow the query to produce optimal output via:

• Description, or "Write a description about this product."
• Follow the example , or "Here are some good product descriptions, write one like this." performance, which model to use, whether it is possible to make a specific tool, and what approach should be taken to do it.

Bernstein said she enjoys the "mad scientist" part of the job, because she can come up with a silly idea and see how it actually works. A lover of poetry, this job role also fueled her fascination with language, a strange intersection of her literary background and analytical mind.

However, this job is unpredictable. New language models emerge all the time, which means Bernstein is always having to readjust her cues. The work itself can be tedious. There were days when she would obsessively modify and test a prompt for hours, sometimes weeks, just to get them to work.

At the same time, there’s no telling what exciting things will happen next. In addition to some people not understanding her work, Bernstein noticed that a big misconception people have about artificial intelligence is that an AI system is sentient when in fact it is not. People may be taken aback when an AI tries to talk, seeing so much of their fears in what it says, but that’s because it’s been trained on people’s fears and sci-fi descriptions of AI.

Tips for writing well are easy to learn but difficult to master. Getting an AI system to do what people want it to do requires trial and error, and over time Bernstein learned to employ some strange strategies, and some of her prompt structures became complex.

Here are some tips that can help develop better AI prompts:

(1) Use thesaurus

Don’t let the first prompt not get you the results you want And give up a concept. In most cases, finding the right words or phrasing will accomplish what you are doing.

(2) Pay attention to the verb

If you want the AI ​​to fully understand the request, make sure the prompt contains a verb that clearly expresses the intent. For example, "Condensate this" is more powerful than "Rewrite this shorter."

(3)ChatGPT is excellent at identifying intent, so use it

Clearly introduce what is to be done from the beginning, and be careful about wording, tense and approach Adjustment. Try writing, "Today, we're going to write about XYZ," or, "We're going to write about XYZ, and we need your input." Trying different approaches to what you're doing can make a difference.

The above is the detailed content of AI tips engineers on three ways to get the best results using ChatGPT. For more information, please follow other related articles on the PHP Chinese website!