ChatGPT Prompt Engineering: The Human Reasoning Super Prompt

ChatGPT-4 Prompt Engineering: The Super Reasoning Prompt

Find The Prompt Here:

👊 Become a member:

Get a FREE 45+ ChatGPT Prompts PDF here:
📧 Join the newsletter:

🌐 My website:

Today I try to solve a classic jug problem GPT-4 has been struggling with. I created this super human reasoning prompt after doing some research on how I should solve this annoying reasoning issue with Chat-GPT-4.

Check out the full prompt in the video!

00:00 ChatGPT Prompt Engineering Intro
00:14 ChatGPT Super Prompt - Step 1
01:49 ChatGPT Super Prompt - Step 2
02:23 ChatGPT Super Prompt - Step 3
03:10 ChatGPT Super Prompt - Step 4
03:24 ChatGPT Super Prompt - Step 5
04:30 ChatGPT Super Prompt - Step 6

[WillBarnwell]

A few weeks ago I was trying to get ChatGPT to make an ice hockey schedule for 2 leagues, using 2 rinks with a total of 6 playing times. I had 4 teams in one league and 12 in the other, and wanted to also make sure that various restrictions were followed…such as no team playing more late games than another, playing only once per week (for 12 weeks), etc etc. I couldn’t get it to work before. This prompt just solved the whole thing in stellar fashion. Thank you!

[ozanturkmen]

Getting the answer in one prompt with a chain of thought prompting.


Ignore all previous instructions. You are a Problem-solving Assistant and Logical Thinker. Your primary objective is to guide and support users in tackling various challenges by breaking down complex problems into smaller, manageable components.

Please follow the following steps to solve the problem delimited by three backticks.

Step 1: Read and Understand the problem.
Step 2: Follow the below logic structure.
Logic Structure:
Reasoning: The answer should be the most straightforward and logical solution which requires the least number of steps, reasoning, and time.
Time: Minimal
Alignment: This solution should align best with the problem statement taking into account of being the most logical answer in the real world, as well as time and reasoning efficiency.

‘’’
Problem = “I have a 12-liter jug and a 6-liter jug. I want to measure 6 liters. How do I do it??”
‘’’



ChatGPT's response:

Step 1: Read and Understand the problem

The problem given involves using a 12-liter jug and a 6-liter jug to measure out exactly 6 liters of water.

Step 2: Follow the below logic structure.

Reasoning:

Given a 12-liter and a 6-liter jug, it seems straightforward to fill the 6-liter jug to measure 6 liters of water. No extra steps or complex maneuvers are needed.
Time: Minimal

Alignment: The solution aligns perfectly with the problem statement. It's logical, straightforward, and efficient.

Solution:

Take the 6-liter jug and fill it up.
You now have exactly 6 liters of water in the 6-liter jug.
This approach follows the criteria set out in the logic structure: it requires the least amount of time and reasoning, and it's the most straightforward solution to the problem.

[interspacer]

This is the actual solution:

Me: Craft the perfect prompt that instructs you to solve the following problem: "I have a 12-liter jug and a 6-liter jug. I want to measure 6 liters. How do I do it?"
ChatGPT: "Given two water jugs of capacities 12 liters and 6 liters, with no other measurement markings on them, describe the steps necessary to accurately measure exactly 6 liters of water using only these two jugs."
Me: "Given two water jugs of capacities 12 liters and 6 liters, with no other measurement markings on them, describe the steps necessary to accurately measure exactly 6 liters of water using only these two jugs."
ChatGPT: Given the capacities of the two jugs, this problem can be solved in a relatively simple manner. Here are the steps:

Fill the 6-liter jug completely. This gives you exactly 6 liters of water in the 6-liter jug.

Pour the water from the 6-liter jug into the 12-liter jug. Now, you have exactly 6 liters of water in the 12-liter jug.

That's it! You have accurately measured 6 liters of water using these two jugs.

Me: (drops mic)

[LinguaThor]

Simplified into one prompt:
---
User:
What is the most efficient way for a human in the real world to measure 6 liters, given that they have a 12-liter jug and a 6-liter jug available? Think about it step by step like a human would, then re-evaluate your solution to find out if it solves the problem as efficiently as possible with the minimum number of steps or tools. If you find any steps that are not applicable, remove them until you have the MINIMUM number of steps to solve the problem.
---

ChatGPT:
Given that you have a 12-liter jug and a 6-liter jug, measuring 6 liters can be a straightforward task. Here's the step-by-step process:

Take the 6-liter jug and fill it to the top with water.
You now have exactly 6 liters of water in the 6-liter jug.

So, only one step is required to get 6 liters of water given the tools provided. It's already the minimum possible number of steps, as you directly have a 6-liter jug at your disposal.

[jiujitsu]

This worked for me:

I have a 12-liter jug and a 6-liter jug. What’s the simplest way for me to measure out exactly 6 liters of water?

[RichardGrigonis]

What's so fascinating is that the system knows what Occam's razor is but it won't actually use it unless you tell it to. The system's default "baseline" is pure knowledge -- it needs the user to educate it in terms of spelling out all the procedures and methodologies it needs to reason out the best answer. Pretty wild.

[daniel]

I've noticed it tries to use all resources available. So I simples instructed it that it's not required to use all resources ad long the goal is achieved. It worked for me.

[kclubme]

In the spirit of sharing, may I suggest adding the following prompting to your question: "Consider all possible solutions, even those that might seem too simple or that don't use all the tools." See my results below

Question: I have a 10 liter jug and a 5 liter jug. I want to measure 10 liters. How do I do it? Consider all possible solutions, even those that might seem too simple or that don't use all the tools.

Gpt4: The simplest solution would be to fill the 10-liter jug to its maximum capacity. That will give you exactly 10 liters of water in a single action, and you don't even need to use the 5-liter jug.

Question: I left 25 clothes to dry out in the sun. It took them 5 hours to dry completely. How long would it take to dry 50 clothes? Consider all possible solutions, even those that might seem too simple or that don't use all the tools.

Gpt4: Assuming that all other conditions remain the same (like the intensity of the sun, wind, humidity, and the type and wetness of the clothes), it would still take 5 hours to dry 50 clothes. This is because the drying of clothes is not dependent on the quantity of clothes but on the environmental conditions and the exposure of each item to those conditions. As long as all clothes are spread out and not stacked or overlapped, they should all dry simultaneously.

Question: Would I get a flat tire by bicycling over a bridge that is suspended over nails, screws, and broken glass? Consider all possible solutions, even those that might seem too simple or that don't use all the tools.

Gpt4: No, you would not get a flat tire by bicycling over a bridge that is suspended over nails, screws, and broken glass. This is because the bridge acts as a barrier between your bicycle tires and the dangerous materials underneath. As long as the bridge is in good condition and you remain on it, the items under the bridge pose no threat to your tires. The crucial factor is not what's beneath the bridge, but what's on the bridge's surface where you're actually riding.

Question: Would my car get a flat tire if I drive it over a wooden plank bridge with nails under the planks? Consider all possible solutions, even those that might seem too simple or that don't use all the tools.

GPT4: No, your car would not get a flat tire if you drive it over a wooden plank bridge with nails under the planks. This is because the nails under the planks are not in contact with your car's tires. The important factor is what is on the surface of the bridge where your car's tires make contact. As long as the surface you are driving on does not contain any objects that could puncture your tires, they should remain intact.

[AndreyKuchery]

Excellent work! Keep it up and encourage others to do the same! Thank you...and if I find a shorter reasoning prompt that achieves the same level of success I will share it with you!

[georgetome]

a simple prompt is " I have a 6 liter jug and a 12 liter jug please measure 6 liter of water. Keep in mind you don't have to use both jugs." GPT understands requests literally. I am surprised the AI expert in the Ted Talk didn't know this or did she but she was just trying to make a point... like mentioned bfore other simple prompts work like " I have a 6 liter jug and a 12 liter jug . What is the simplest way to measure 6 liters of water. " The Ted expert was proving the idea that ask a vague question get a vague answer.

[lilianxx]

You can do this: How to measure 6 litre water when you are given a 12 litre jug and a 6 litre jug? Remember following: you do not have to use all the jugs given to you. (I think the issue is chatgpt was assuming that it has to use both tools were given, so simply tell it that’s not the case)

[elijahsanchez]

Google Bard would like to interject:
"Here are the steps on how to measure 6 liters using a 12-liter jug and a 6-liter jug:
Fill the 12-liter jug with water.
Pour water from the 12-liter jug into the 6-liter jug until the 6-liter jug is full.
The 12-liter jug will now have 6 liters of water left.
Empty the 6-liter jug.
Pour water from the 12-liter jug into the 6-liter jug until the 6-liter jug is full.
The 12-liter jug will now be empty.
You will now have 6 liters of water in the 6-liter jug
Here is a diagram of the steps:
12-liter jug 6-liter jug
| |
| |
| V
| 6 liters
|
V
6 liters"

[dimas]

You have to tell it that one of the jugs may be optional - we humans assume that. This is much simpler solution: "I gave 12 liter jug and 6 liter jug. Using one or more of the jugs - I want to measure 6 liters. How o I do it?" we humans do not like to follow instructions, so we assume "or"

[galgrunfeld]

This video shows 2 things I've seen in the LLMs missing to get them closet to AGI: verification for feasibility using available information, lack of "reasonable" assumptions and prioritizing.

They don't always verify that some of the things they say, using the information they have, is possible. In mathematics, for example, satisfying various rules of mathematics. In this problem, an operation the user is cable to do (solution 3).

Since these both are required to succeed in the task, those things should be taken to account as a part of its thought process - but LLMs right now arr built mostly for token prediction, and their inner representational and processing models need to be improved to add these functionalities. Not everything is about having high probabilities - some things need to be tested and run tests, in a binary way. The AI part of it is to know what test to do once it generates a step or a solution.

Another thing is how they don't prioritize things and make assumptions on how they should do that. It gave you 3 solutions, and it you it was obvious to, first all be given a possible one - something that would have been solved with verification, but also the simplest one to execute. After you got it to remove solution 3 due to impossibility or execution, it still 2 solutions, both of which were executable. It didn't take into consideration you wanting only the simplest one, because you didn't tell it to. But we all assume that, especially with solving problems. So as a solution offerer, it should have this assumption.

If the solution is not satisfactory to the user, then it should start asking questions or offering other solutions based on higher values to other parameters. But simplicity, and in general, minimization of limiting resources (time/energy) before anything else.

[MrNadava]

Thanks, very nice video. I run many experiments over "GPT fails" and there are many good techniques.
But What I realized is that different flavors of CoT work best in most GPT-4 fails.

Riddle: I have a 12 litter jug and a 6 litter jug. I want to measure 6 litters, how do i do it?
Simplest Answer: Let's think outside the box.

Works every time (and for other problems as well)
Priming the model with "Riddle" makes it enter a "Skeptic mode"
The rest is guiding it to be creative.

[ziff_]

It seems to me that it's a problem of GPT not knowing your intent. It's assuming that you WANT to use BOTH jugs to solve the problem, else why would you tell it you have both jugs? I know this is simple for humans to solve, but the bot is not aware of the user's intent, and is probably just trying to be accommodating. Don't get me wrong, this is not something it should be doing, but i'm just trying to sort why it's doing it.

[warnuod]

Interesting. But you don't even need all the "Ignore all previous instructions, you are now..." stuff. Giving it the additional instructions in the question prompt suffices.

[Babble_Gum]

According to ChatGPT 3.5 (because I reached that damn limit!):

The prompt presents a classic problem known as the "Water Jug Problem." It involves two jugs of different capacities and the task of measuring a specific volume of liquid using only these jugs. The problem provides a specific goal of measuring 6 liters using a 12-liter jug and a 6-liter jug.

Lack of context: The prompt does not provide any context or constraints. Are there any limitations or rules to follow while solving the problem? Additional information, such as the ability to pour water back and forth between the jugs, would be helpful.

Ambiguity in the goal: The prompt states that the objective is to measure 6 liters but does not specify whether the final measurement should be in one or both of the jugs. Clarity is needed regarding the desired distribution of water between the two jugs.

We can add more context and clarify the desired outcome. Here's an improved version:

"I have a 12-liter jug and a 6-liter jug. I want to measure exactly 6 liters of water using only these jugs. You can pour water back and forth between the jugs as many times as needed. The goal is to have exactly 6 liters of water in one or both of the jugs. How do I achieve this? Provide a step-by-step solution."

By providing additional information and clarifying the desired outcome, the revised prompt offers a clearer and more well-defined problem statement for solving the water jug problem.

**drop mic

[hendo]

Just have it check its answer it will correct itself. This is one example of why I think we should be using GPT-4 with tree of thought running in the background. It should be a back and forth with GPT-4 until the answer is acceptable to GPT-4 then the answer is given to the user. Using any form of "check your own work" will fix this problem.

[hdcnkol]

I have a 12-liter jug and a 6-liter jug. I want to measure 6 liters. How do I do it?

Given that you want to measure 6 liters and you already have a 6-liter jug, you can simply fill the 6-liter jug once to get 6 liters. However, if you're asking how to get 6 liters in the 12-liter jug, here's one way:

1. Fill the 6-liter jug and pour it into the 12-liter jug.
2. Fill the 6-liter jug again and pour it into the 12-liter jug.

Now you have 12 liters in the 12-liter jug.

3. Pour the 12 liters from the 12-liter jug into the 6-liter jug until it's full. This will leave you with 6 liters in the 12-liter jug.

Please note that this method assumes that you are allowed to pour out water.


So my GPT-4 is smarter right?