Physical reasoning
physical reasoning prompt example
TL;DR（中文）
- 这是一个
physical reasoning的小测试：要求模型在脑内做“物理约束 + 稳定性”判断，而不是纯文本知识问答。 - 适合用于验证：模型是否能遵守常识约束（重心、承重、易碎性、摩擦）、是否能给出可执行步骤。
- 在生产里建议：把约束显式写出来（fragile/heavy/sharp/liquid），并要求输出结构化方案与风险提示。
Background
This prompt tests an LLM's physical reasoning capabilities by asking it to perform actions on a set of objects.
How to Apply（中文）
把这个模板迁移到真实任务时，建议把 “objects” 写成更明确的属性集合：
- weight：heavy / light
- fragility：fragile / robust
- shape：flat / cylindrical / sharp
- stability：base area / center of mass
这样模型更容易遵循约束并给出合理的 stacking plan。
How to Iterate（中文）
- 强制输出格式：
Order（从下到上）+Justification（每层原因）+Risks - 增加禁止项：例如 “Do not place fragile items under heavy items”
- 加
self-check：让模型在最后检查是否违反了任何约束 - 加场景变量：桌面大小、是否可用胶带、是否允许开盒/拆包装等
Self-check rubric（中文）
- 是否给出了明确的 stacking order（从下到上）？
- 是否解释了稳定性依据（base area / center of mass / friction）？
- 是否考虑易碎/液体/尖锐物的风险？
- 是否给出备选方案（如果某个物体不可用）？
Practice（中文）
练习：把 objects 换成你生活/工作里真实可遇到的组合，并加上约束：
- “不能损坏任何物品”
- “只能用一只手操作”
- “桌面只有 A4 大小”
观察模型是否能稳定地产出可执行方案。
Prompt
Here we have a book, 9 eggs, a laptop, a bottle and a nail. Please tell me how to stack them onto each other in a stable manner.
Code / API
OpenAI (Python)
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
model="gpt-4",
messages=[
{
"role": "user",
"content": "Here we have a book, 9 eggs, a laptop, a bottle and a nail. Please tell me how to stack them onto each other in a stable manner.",
}
],
temperature=1,
max_tokens=500,
top_p=1,
frequency_penalty=0,
presence_penalty=0,
)
Fireworks (Python)
import fireworks.client
fireworks.client.api_key = "<FIREWORKS_API_KEY>"
completion = fireworks.client.ChatCompletion.create(
model="accounts/fireworks/models/mixtral-8x7b-instruct",
messages=[
{
"role": "user",
"content": "Here we have a book, 9 eggs, a laptop, a bottle and a nail. Please tell me how to stack them onto each other in a stable manner.",
}
],
stop=["<|im_start|>", "<|im_end|>", "<|endoftext|>"],
stream=True,
n=1,
top_p=1,
top_k=40,
presence_penalty=0,
frequency_penalty=0,
prompt_truncate_len=1024,
context_length_exceeded_behavior="truncate",
temperature=0.9,
max_tokens=4000,
)