Shulou(Shulou.com)11/24 Report--

According to news on April 13, the recently popular chat robot ChatGPT can provide very high-quality answers, but only if you know how to ask it the right questions. So how do we enter the correct prompt (prompt)? The American media have provided many proven and effective techniques.

The following is the translation:

Today, we can use generative artificial intelligence tools, such as ChatGPT, to get the answer by typing in a few prompts. However, this does not mean that these answers are always accurate. The key point is that you need to enter the prompt information correctly.

Can you use simple words to communicate with the latest artificial intelligence systems? Peter Lin, a professor of computer science at the University of Waterloo, said the idea was "very deceptive".

Today, many companies are recruiting so-called cue engineers who have the skills and experience to communicate effectively with large language artificial intelligence systems. Many job seekers also want to master this skill.

According to professional social networking site LinkedIn, the number of global members using terms such as ChatGPT, prompt engineering and generative AI in February increased by 55% compared with January and 71% in March compared with February.

"We now have 'Dog Whisperer' and 'Horse Whisperer'," said Peter Lin. "now we are developing whispers for large language models."

In general, the more detailed the prompt that the user enters, the closer the response given by ChatGPT will be to the result the user wants. But the most effective hint is not just to add details.

Figure 1: for better tips, you can add specific formatting requirements, such as writing style, list, or summary. Here is an example:

Basic tip: "what is ETF?"

A better tip: "how do you explain ETF to a layman? can you give me three examples to help me understand it?"

First, in order to get the best response, it is important to add the right details. Moreover, there are many ways to guide an artificial intelligence system to present the information it finds, depending on the goal. For example, you can tell it to provide answers in a certain style or for a specific audience. Each tool also has specific features and limitations that require users to learn how to use it. The best results usually come from a long conversation with artificial intelligence, responding to what it produces and asking for improvement.

Brian Kimmel, founder of venture capital fund Worklife, says that even in applications that do not explicitly mention "prompt project" jobs, having this skill can become a must-have for multiple positions. For example, art directors can use generative artificial intelligence to outline ideas, marketers can use it to create new slogans, software engineers can use it to find problems in code, and lawyers can use it to study legal provisions.

Worklife is a venture fund that invests in companies that use generative artificial intelligence.

"it's going to be a basic skill that all of us need to know," Kimmel said. "because once you learn a new skill, you can apply it anywhere you like."

Figure 2: basic tip: "write a love poem for me and give it to my partner."

Better tip: "write a love poem for my mysterious partner. His favorite poem is" the Crow "by Edgar Edgar Allan Poe."

Basic tip: "make a five-day vegetarian plan."

A better tip: "prepare vegetarian recipes and make a shopping list for the next five days."

Essential job skills Sophie Sophie Antebi, who will graduate from the University of Southern California in May, will have degrees in marketing and entrepreneurship and will be interested in working for startups. In order to make her resume stand out, she learned to use a variety of generative artificial intelligence tools. Anteby believes these tools can be useful when planning marketing campaigns or creating images for presentations.

Although Anteby has not yet found a job, she says her application has received more responses since she added "prompt engineering" and other artificial intelligence skills she is familiar with in her cover letter. She believes that these skills helped her successfully enter the second and third round of interviews. Speaking of his interviewer, Anteby said: "they ask a lot of relevant questions and are very interested in these skills."

Boston Children's Hospital is looking for a tip engineer. The hospital released a recruitment message in February and is currently interviewing candidates. John Brownstein, the hospital's chief innovation officer, said it prompted engineers to work in the hospital's innovation and digital health accelerator division, which has continued to invest in artificial intelligence over the past few years.

Dr. Brownstein said that large language artificial intelligence tools will play a role in many work in hospitals, which may include analyzing a large number of unstructured clinical records, detecting medical rules that might have been ignored, and providing patients with health knowledge. and improve the efficiency of patient admission. He added: "artificial intelligence technology like ChatGPT must have a hype, but that doesn't prevent us from adopting it."

Nick West, chief strategy officer of London-based law firm Mishcon de Reya, which also plans to hire AI prompt engineers, says large language AI tools are most likely to be used in litigation as the first step in summarizing legal documents, finding relevant case law and querying databases of information.

"I believe this is a transformative technology," West said. "so any help can accelerate our understanding and understanding of it, which means we can use it faster."

Elvis Saravia, co-founder of Dair.ai, says that even without full-time AI prompting engineers, the technical team needs to be familiar with artificial intelligence tools. This situation is becoming more and more obvious. Dair.ai is an organization dedicated to popularizing artificial intelligence tools through open source research, educational materials, and training. Saravia quickly released an engineering guide that instructs engineers on how to build and dock large language models.

Figure 3: basic question: "make a plan to stay in Paris for three days."

A better tip: "make a complete plan for travelers to stay in Paris for three days, including staying at hotels that cost less than $500 per night and provide gluten-free dining places."

Collaborative art researchers and users have discovered and documented many tips. One is to tell the large language model "Let's think about it step by step", which allows the technology to break down complex requests into smaller problems that are easier to deal with. Another way is to tell artificial intelligence that it is an expert in a certain field and its job is to provide advice to laymen. This helps to simplify the answer to complex problems.

Some researchers and users are posting guidelines and tips they have found. Some users even sell their prompts on the market, while others are developing tools that can optimize prompts. But many resources are shared for free on Twitter posts and YouTube videos. Includes resources released by Nick Pierre (Nick St.Pierre), a designer who created content focused on interpreting Midjourney and other artificial intelligence tools for artistic creation.

For works of art, we need to combine the experience of using tools and the knowledge of artistic style to improve the tips. Pierre suggested starting with a basic concept and then replenishing it as needed. He said, for example, that artificial intelligence might be asked at first to provide a picture of a 40-year-old dark-haired woman sitting in a bar. If the final image is not what the user wants, they can add terms to change the camera angle, clothing style, lighting and other factors to get different results.

Pierre found that the leading factor in the prompt was usually more prominent in the final image. So he tends to list what he thinks is most important in visual effects first. "We need to iterate and test many times and work closely with artificial intelligence tools," he said. "

How to give a better hint on ChatGPT? 1) understand the functions of ChatGPT

Although ChatGPT can do many things, it is not omnipotent. For example, it is just a language model and is not good at math. In addition, ChatGPT is currently only trained until 2021, so any events after that are unknown to it.

OpenAI didn't say what data it used to train ChatGPT, but Andrew Mayne, the company's science communicator, advises people to read what other people are doing on ChatGPT and where they have been successful. He also encourages people to try it for themselves in order to figure out how to make it do what people want it to do reliably.

2) make sure the prompt is accurate and clear

Inaccurate words can confuse ChatGPT, just as it can confuse people. Ambiguous wording, typos and omissions of words can all lead to such a result. Homonyms can be particularly troublesome: in the absence of a clear context, artificial intelligence may be able to recognize words but produce ambiguity. For example, "Right" can indicate both direction and "correct".

3) adjust the prompt

ChatGPT will refine its response based on new input from the user, from the request to a specific description of the format. For example, users can ask it to show its response in the form of a summary, or try to give an answer in a more serious or well-known writer's tone, such as Mark Twain.

4) ask for help

If the initial query is unsuccessful, you can ask ChatGPT to help you design better tips. For example, you can say, "what do I need to tell ChatGPT to write a good cover letter for me?" This artificial intelligence tool will tell you the information it needs.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.