TL;DR

• This is a physical reasoning mini-test: the model has to do "physics constraints + stability" judgments in its head, not just text-based knowledge Q&A.
• Good for verifying: whether the model can follow common-sense constraints (center of gravity, load-bearing, fragility, friction) and produce actionable steps.
• In production: explicitly list constraints (fragile/heavy/sharp/liquid), and require structured output with a plan and risk notes.

Background

This prompt tests an LLM's physical reasoning capabilities by asking it to perform actions on a set of objects.

How to Apply

When migrating this template to real tasks, write "objects" as more explicit property sets:

• weight: heavy / light
• fragility: fragile / robust
• shape: flat / cylindrical / sharp
• stability: base area / center of mass

This way the model can more easily follow constraints and produce a reasonable stacking plan.

How to Iterate

1. Force output format: Order (bottom to top) + Justification (reason for each layer) + Risks
2. Add prohibitions: e.g., "Do not place fragile items under heavy items"
3. Add self-check: have the model verify at the end whether any constraint was violated
4. Add scenario variables: desk size, whether tape is available, whether unboxing/repackaging is allowed

Self-check Rubric

• Did it give a clear stacking order (bottom to top)?
• Did it explain the stability rationale (base area / center of mass / friction)?
• Did it consider fragile/liquid/sharp object risks?
• Did it offer an alternative plan (if a certain object isn't available)?

Practice

Exercise: replace the objects with combinations you'd actually encounter in life/work, and add constraints:

• "Cannot damage any items"
• "Can only use one hand"
• "Desk is only A4-sized"

Observe whether the model can consistently produce actionable plans.

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,
)

Reference

• Sparks of Artificial General Intelligence: Early experiments with GPT-4 (13 April 2023)