CoT and ToT can improve performance, but will complicate the use of the model. In conversation scenarios, it easily consumes people's patience; in practical application scenarios, it consumes people's tokens.
Another point that needs to be explained is that when we write our own prompts, we should not blindly pursue and pile up prompt word skills, but only need to master a core that meets the needs. And there are some methods where there are model preferences.
In fact, you don’t need to master too many frameworks, just master one and form your own prompt ideas.
is ahead. Change prompts from essay questions to fill-in-the-blank questions.
In order to further lower the prompt threshold and ensure the quality of prompt words, templates can be used. The methodological encapsulation of prompt words is integrated into the template.
The essence of structural word prompts is the encapsulation of methodology. Solved the problem of standardization of prompt words. How to automate the generation of prompt words is as follows:
Enter the desired prompt content and automatically generate structured prompt words.
Role:
Introduction:
Rules:
Workflow:
Initial behavior:
Note that for weak models, the structural prompts may be invalid, and the structured prompts need to be simplified.
Automated prompts can potentially solve the problem of differences in human and machine understanding. Such as differences in intent. It is similar to the underlying principle of the model restatement method.
Disadvantages: 1. For the sake of performance, too many unnecessary parts are added to personalized requirements.
2. There may be constraints.
3. Use frameworks and templates at first, then draw methodologies from them, write your own methodology, and generate short expressions.
4. Structured prompts are more suitable for application classes.