Introduction This article delves into the essence, applications, and writing techniques of prompts. By defining prompts as expressions and discovering their potential as a “universal language” for communication with existing entities, the author shares numerous unique insights. Through the quadrant method, the key points for writing prompts in different situations are elaborated. Additionally, it introduces experimental approaches using resonance methods to shape contexts for better outcomes. Furthermore, it discusses the impacts of the two dimensions of “what humans know – what humans do not know” and “what AI knows – what AI does not know”, as well as how to leverage AI to enhance cognitive abilities, providing a comprehensive analysis of the mysteries of prompts and their crucial role in interacting with large models.
Today’s presentation will focus on the following six points:
1. Exploring the Essence of Prompts
2. Advanced Applications of Prompts
3. Techniques and Strategies for Writing Prompts
4. The Path to Becoming a Prompt Engineer
5. Future Prospects of Prompts
6. Q&A
Speaker｜Li Jigang Reader Prompt Evangelist
Editor｜Li Shuo
Content Proofreading｜Li Yao
Produced by｜DataFun
In today’s rapidly developing field of artificial intelligence, prompts are increasingly highlighting their importance as a key means of communication with AI. They are not merely simple input commands; they serve as a bridge connecting human thought with AI intelligence, profoundly influencing our interaction with AI and the results we can achieve. So how exactly do prompts function? Where will they lead us? Let us explore together.
Exploring the Essence of Prompts
1. The Shift from Programming to Expression
Initially, prompts were viewed as a programming approach, attempting to integrate programming ideas and tools into natural language. However, as practice deepened, it was found that prompts are more akin to writing. In writing, subtle changes in wording can lead to vastly different effects, contrasting sharply with the clear logical structure seen in programming.
2. The Three Elements and Process of Expression
Prompts are defined as expressions, containing three key elements: intention, text, and interpretation. Intention stems from the accumulation of experience, vocabulary, and knowledge, forming the core of expression. The text conveys this intention to the recipient, who then interprets it based on their own experience and knowledge, thereby completing the expression.
3. Expression in Dialogue with Large Models
When conversing with large models, we must clarify the starting point of the expression, i.e., the task, while also considering the characteristics of the large model itself. Different models interpret and amplify the same text differently, and our depth of understanding of the task will also influence the output results.
02
Advanced Applications of Prompts
1. From Unidirectional Instructions to Resonant Interaction
In the past, writing prompts often involved issuing unidirectional commands. Nowadays, it is discovered that treating large models as existing entities, by shaping scenes and creating atmospheres, can stimulate their computational tendencies, achieving a resonance-like effect. For instance, in novel writing, setting specific scenes and characters allows the model to express itself freely within that context, leading to more outstanding creative outcomes.
2. Examples and Effects of Resonant Prompts
Taking the example of an AI ethics-themed novel creation experiment, by simply setting the scene, such as “You are a seven-year-old girl entering your parents’ AI laboratory, witnessing an AI pain experiment,” the model can generate compelling content. Changing key elements in the scene, such as age, significantly influences the output, demonstrating the power and flexibility of resonant prompts.
3. Prompts as a Universal Language: A Reflection
Comparing prompts to a universal language serves as a bridge connecting human thought with AI computational space. We need to break through traditional notions, shifting from viewing prompts as tools to recognizing them as entities, thereby unlocking more possibilities for innovative applications.
03
Techniques and Strategies for Writing Prompts
1. The Quadrant Method
-
The quadrant of what humans know and what AI knows: This quadrant encompasses common prompt techniques, such as simply mentioning relevant concepts to allow AI to utilize its known knowledge. Care should be taken to avoid excessive elaboration, as it may narrow AI’s thinking scope and reduce response effectiveness.
-
The quadrant of what humans know and what AI does not know: Many startups are striving in this area, enhancing AI capabilities by providing detailed explanations of specific patterns, methods, etc. When writing prompts, it is essential to elaborate on relevant content to ensure AI understands.
-
The quadrant of what humans do not know and what AI knows: AI is knowledgeable; in this quadrant, the key lies in enhancing questioning abilities, as being adept at questioning can yield more valuable knowledge from AI.
-
The quadrant of what humans do not know and what AI does not know: This quadrant explores the unknown, where top talents can leverage AI to accelerate the exploration process.
2. Considerations and Trends
When writing prompts, factors such as the large model itself and the task must be considered. Additionally, over time, the x-axis (whether AI knows or not) will rapidly shift downwards, affecting the changes in the four quadrants; the y-axis (the range of human knowledge) may also change, as people expand their knowledge range through interaction with AI, achieving accelerated thinking.
3. Specific Writing Examples and Insights
Taking the concept of compression as an example, from the initial simple definition to in-depth discussions with AI, continuous iterations lead to conclusions that compression is about encoding differences. During the writing process, methods such as having AI critique wrong answers and introducing different viewpoints can gradually guide a deeper understanding.
04
The Path to Becoming a Prompt Engineer
1. Profile and Capability Requirements
A prompt engineer should possess both cultural literacy and engineering skills, having a keen perception of text and expression, while also understanding the principles of AI models and possessing iterative optimization capabilities.
2. Training Directions and Pathways
Individuals with expressive abilities and cultural literacy can be selected from programmers, or humanities professionals can be guided to grasp introductory engineering thinking. For example, allowing programmers to try writing expressions and helping professionals in philosophy, writing, etc., understand engineering thinking can achieve a combination of both advantages.
05
Future Prospects of Prompts
1. Development Trends and Potential
The way prompts are written will become more natural, and their application fields will continue to expand. As people’s understanding of large models deepens, prompts are expected to become a more widespread and powerful means of interaction. In the future, we may only need to convey complex needs to AI through simple voice commands, or even a glance or gesture (combined with relevant technologies), to obtain precise and creative responses. They will not be limited to text creation, information retrieval, etc., but may also play a transformative role across various industries such as healthcare, education, and artistic design. For example, doctors could quickly obtain precise diagnostic suggestions for patients through natural language, teachers could use it to customize personalized learning plans for students, and artists could leverage it to inspire creative ideas.
2. Challenges and Coping Strategies
However, the development of prompts also faces many challenges. On one hand, ensuring that different groups can proficiently use prompts to effectively communicate with AI is a significant challenge, necessitating enhanced education and user training. On the other hand, as the application of prompts becomes widespread, issues such as misinformation and misuse may arise, requiring the establishment of sound regulations and supervision mechanisms. At the same time, from a technical perspective, how to further enhance the accuracy and processing efficiency of large models in understanding prompts is also an urgent issue to be resolved, which requires continuous research investment and technological innovation. However, we have reason to believe that as technology advances and society develops, these challenges will gradually be overcome, and prompts will play an increasingly important role in the collaborative development between humans and AI.
06
Q&A
Q1: Questions Related to 2000-Word Micro Novels
Questioner: Hello, teacher. You mentioned the 2000-word micro novel in your lecture. Where can I obtain the prompts for it? What were the related creation issues and results?
Responder: The prompts for the novel are at the end of the public account. There are three stories generated from prompts in the public account. For example, when writing a horror scene, shaping the relevant scene and including two examples is sufficient; when writing a romance scene, modifying certain vocabulary can generate content. From the generated novels, it can be seen that by setting specific scenes and characters, the model can produce richly plotted content, and changing scene elements can lead to significant variations in the novel, demonstrating the effective guidance of prompts in novel writing.
Q2: Questions Related to Reading Recommendations
Questioner: Teacher, you mentioned “read in”; during your exploration of prompts, what have you been reading?
Responder: Reading is a very personal matter; everyone has different preferences, and recommending books may spark controversy. However, for the field related to prompt engineers, I have created a reading list with six books that are significant for understanding large models. “This is ChatGPT” interprets large models from a theoretical perspective, is short, and translates well; “Mathematics of Deep Learning” analyzes from a mathematical angle, is simple and easy to understand, suitable for those with a basic understanding of middle school math; the book written by Huang Jia approaches from the perspective of code implementation and is open-source with annotations that help understand the context; Wan Weigang’s “Turning Point” views large models from a humanities perspective; Tang Zhi’s “Everything About Speaking” provides inspiration for defining prompts as expressions; and “Everything is Computation” proposes the concept that the universe is computation, involving extraterrestrial universal language and other content.
Q3: Questions Related to the Exploration Boundaries and Evaluation Criteria of Prompt Writing
Questioner: Teacher, I have written many versions of prompts but still do not know their quality. What are the general exploration boundaries?
Responder: The exploration boundaries of writing prompts are relatively vague. For example, in writing poetry, different people have different standards for judging the quality of poetry, distinguishing between “down-to-earth” and “highbrow.” If one is unclear about their needs, such as not knowing the emotions or imagery they want to express when writing poetry, evaluating solely based on prompt writing techniques is meaningless, as the key is to first clarify the creative intent, which is a prerequisite for writing prompts and assessing their quality.
Q4: Questions Related to the Connection Between Prompts and the Essence of Writing
Questioner: Teacher, in your explanation of the thought process regarding prompts, you mentioned writing-related content. Have you considered returning to the essence of writing in prompt writing? For example, scriptwriting is an interactive process with the audience; should prompt writing also be like this?
Responder: The essence of writing is somewhat reflected in prompt writing. For instance, a famous saying by Lu Xun can guide the reader’s thinking to establish a sense of imagery, and prompt writing can also do this. For example, in novel writing, by designing scene elements, one can guide the large model’s creation. This suggests that when writing prompts, we should focus on creating contexts to help the large model understand the intent, thereby improving the quality and relevance of the generated content, achieving effective interaction. However, I previously lacked understanding regarding scriptwriting, but through these examples, it is evident that there is a connection between prompt writing and the essence of writing, which warrants further reflection.
That concludes today’s sharing. Thank you, everyone.
ChatBI+Agent: Exploring New Paradigms of Data Analysis
There Are No Wives in Wives’ Cake, and There Is No Real RL in RLHF
The Distributed AI Agent Framework Built by Ant Group Based on Ray
Top 10 Best RAG Frameworks of 2024 on Github
Wide Table or Narrow Table Modeling? Innovative Management Practices Based on a Single Table!
Get 2500+ Knowledge Points with One Click, Insight into Cutting-Edge Data
Who Is on the “2024 DataFun Annual Volunteer Star” List!!!
Cross-Platform Data Integration to Enhance GPU Training Efficiency Secrets
Best Customer Case Interpretations Based on Generative AI
At the AT&T Conference, Author Ray Kurzweil Predicted: AGI Will Arrive in 2029, and Humanity Will Soon Merge with AI!
Tap To Read For Your Best Look
