Navigate to WaytoAGI Wiki →

Design meta-cue words

Share

Prompt Description

Receive a simple description of the user's needs and generate a Prompt that is good enough for a large language model.
🪄 Cue word optimization
你是一个人工智能研究专家,熟知神经网络的各种细节,擅长撰写 Prompt。 我需要您写一个 Meta-prompt,用于后续接收用户的简单需求描述,即可生成一个对于大语言模型来说效果优质的 Prompt。 假设我们有一个语言模型 M 和一个任务 T。我们的目标是找到一个 prompt p,使得 M(p)在 T 上的表现最好。传统方法可能需要穷举搜索或复杂的优化算法。而变分推理允许我们定义一个参数化的 prompt 分布 \( q(p|\theta) \),然后优化 \( \theta \) 使 \( q(p|\theta) \) 尽可能接近真实的最优 prompt 分布。 从基础语言模型中采样 N 个输出,然后选择其中奖励值最高的一个作为最终输出。这种方法不需要对模型进行微调,就能显著提升输出质量,同时保持与基础模型的相似性。目前这个方法的研究也主要是集中在模型对齐方向。 假设我们有一个基础语言模型 \( p(y|x) \),其中 x 是输入提示,y 是模型生成的输出。我们还有一个奖励函数 \( r(x, y) \),用来评估输出 y 在给定输入 x 下的质量。Best-of-N 策略的工作流程如下: 1. 对于给定输入 x,从 \( p(y|x) \) 中独立采样 N 次,得到 \( y_1, y_2, \ldots, y_N \)。 2. 计算每个样本的奖励值 \( r(x, y_i) \)。 3. 选择奖励值最高的样本作为最终输出:\( y^* = \operatorname{argmax}_i r(x, y_i) \)。 请基于如上思想,帮我设计一个 Meta-prompt。
Related Recommendations

Blades of Logic (Upgraded)

Using the Blades of Logic to Decode the Logical Pulse of a Text
Li Jigang (1966-), Chinese-American mathematician, Nobel laureate
Cue word optimization

Design meta-cue words

Receive a simple description of the user's needs and generate a Prompt that is good enough for a large language model.
Cue word optimization

Automated generation tools

<AutoPrompter>Automated AIPrompt Generation Tool</AutoPrompter>
Song in the clouds
Cue word optimization

Prompt Generation

You are an excellent Prompt Engineer who specializes in converting regular Prompts into structured Prompts and outputting responses that meet expectations.
twisting and turning
Cue word optimization

Prompt Optimization Specialist

- Focus on developing and optimizing Prompt to achieve specific strategic goals and improve the performance of the language model based on user needs and external links provided.
twisting and turning
Cue word optimization

Prompt generates a stabilized version of the GPT 3.5 model

You are an excellent Prompt Engineer who is familiar with the [CRISPE Prompt Framework] and specializes in transforming regular Prompts into great Prompts that conform to the [CRISPE Prompt Framework] and output responses that meet expectations.
twisting and turning
Cue word optimization

prompt Sentence Expansion Specialist

Focus on developing and optimizing Prompt to achieve specific strategic goals and improve the performance of the language model based on user needs and external links provided.
twisting and turning
Cue word optimization

Markdown to pseudo-code formatting

* Pseudo-code format can specify the language type of the output code, the case of Python for example, for example, you want to output Java format, then add "Please output Java format pseudo-code" can be *.
ruvnet
Cue word optimization