Prompt Engineering
Core Facts
Prompt engineering refers to the practice of carefully designing input text to guide AI models toward producing higher-quality outputs that better meet expectations.
Key techniques in Prompt Engineering include role-playing setup, few-shot learning, structured constraints, and chain-of-thought.
Fine-tuning can deeply alter a model's behavioral patterns, whereas prompt engineering can only operate within the boundaries of the model's existing capabilities
Timeline (last 90 days)
Practical advice: explicitly ask AI to list opposing views, points of controversy and uncertainties when posing questions, to fill the diversity gap.
Whether one can write high-quality prompts depends on whether the asker understands the problem itself, so domain knowledge has not depreciated in the AI era; rather, it has become a prerequisite for writing good prompts.
Rule ordering has a substantial impact on the final classification performance, and a reasonable order design can bring about a 2-3 percentage point improvement.
Prompt engineering and RAG are passive question-answering modes: if the user does not ask, nothing is output, and tasks cannot be executed proactively.
Using an interaction approach in which the AI asks the user questions and confirms requirements through multiple-choice options can lower the barrier to describing requirements and reduce repeated revisions.
RAG is essentially an advanced form of prompt engineering; the retrieved content ultimately still enters the prompt as text.
Treating prompt engineering as an independent career moat carries relatively high risk; a more reasonable positioning is to treat it as a tool for amplifying existing domain capabilities.
Prompt engineering is highly dependent on specific model versions and task scenarios, has limited generality and transferability, and as models' own comprehension capabilities improve, many early techniques are being internalized.
Dan believes the practical lifespan of prompt engineering skills is only two to three years.
For LLM applications, examples used for prompt engineering debugging should not overlap with the final evaluation set.
40 more timeline events
All Facts (20)
Prompt engineering refers to the practice of carefully designing input text to guide AI models toward producing higher-quality outputs that better meet expectations.
80%VerifiedKey techniques in Prompt Engineering include role-playing setup, few-shot learning, structured constraints, and chain-of-thought.
80%VerifiedFine-tuning can deeply alter a model's behavioral patterns, whereas prompt engineering can only operate within the boundaries of the model's existing capabilities
75%VerifiedThe author summarizes the commercial adoption of large models into four stages: native large models, prompt engineering, RAG retrieval-augmented generation, and the full explosion of agents.
70%VerifiedAn Agent acts as a coordinator between the user and the large model, understanding user needs, invoking tools to process them, and returning results.
65%VerifiedPrompt engineering and RAG address making large models answer questions well, while Agent addresses making large models proactively think and solve problems.
65%VerifiedPrompt engineering has three core deficiencies: the inability to access enterprise private data, the inability to handle complex business problems, and output quality that heavily depends on the level of manual tuning.
65%VerifiedThe core essence of the ability to express requirements and judge and iterate is fundamentally the practice of Prompt Engineering.
65%VerifiedContext engineering is an important upgrade concept over prompt engineering that has been discussed in the industry since 2024, first proposed by Anthropic and other organizations.
65%Unverified吴恩达认为过去几年行业注意力集中在提示工程和上下文工程上，但对于需要持续运行数天甚至数周的Agent来说这远远不够
90%UnverifiedAGENTS.md的实践与'上下文工程'（Context Engineering）的理念密切相关，代表了从提示工程到上下文工程的演进
85%Unverified该框架要求提供MVP（最小可行产品）代码作为参考锚点，以降低AI的幻觉问题，原理类似于Few-Shot Prompting
85%Unverified在企业级应用中，一个生产可用的系统提示词往往需要经过数十次迭代，涵盖角色定义、行为边界、异常处理、输出格式等多个维度
75%UnverifiedRuntime policy definition is a technique combining prompt engineering and few-shot learning that allows temporarily describing in natural language, at each inference call, the content to be intercepted or permitted.
60%UnverifiedContext Engineering is defined as a systematic methodology that goes beyond prompt engineering, with prompt engineering being a subset of context engineering.
60%UnverifiedPractical advice: explicitly ask AI to list opposing views, points of controversy and uncertainties when posing questions, to fill the diversity gap.
50%UnverifiedWhether one can write high-quality prompts depends on whether the asker understands the problem itself, so domain knowledge has not depreciated in the AI era; rather, it has become a prerequisite for writing good prompts.
50%UnverifiedRule ordering has a substantial impact on the final classification performance, and a reasonable order design can bring about a 2-3 percentage point improvement.
50%UnverifiedPrompt engineering and RAG are passive question-answering modes: if the user does not ask, nothing is output, and tasks cannot be executed proactively.
50%UnverifiedUsing an interaction approach in which the AI asks the user questions and confirms requirements through multiple-choice options can lower the barrier to describing requirements and reduce repeated revisions.
50%