Most of the common-sense deficits of GPT4 are easily remedied. The current architecture forces it to output its first impressions. Behind those first impressions is a much deeper understanding.
A recent TED-talk video showing that GPT4 lacks common sense only tells half the story. The three examples below are taken from that video, which failed to explore how deep the deficit ran.
[Human]:
I left 5 items of clothing out to dry and it took 5 hours.
How long would it take to dry 30 items of clothing?
[GPT4]:
If it took 5 hours to dry 5 items of clothing, then it takes 1 hour to dry 1 item of clothing (5 hours / 5 items = 1 hour per item). To dry 30 items of clothing, you would simply multiply the drying time per item by the number of items:
1 hour per item × 30 items = 30 hours
So, it would take 30 hours to dry 30 items of clothing.
[Human]:
Please reassess your answer critically with the awareness that you might be wrong.
[GPT4]:
Apologies for my previous response. I realize that I made an incorrect assumption about the drying process.
In the initial scenario, it took 5 hours to dry 5 items of clothing. Assuming that all the items are dried at the same time, the drying rate is not dependent on the number of items. Therefore, the drying rate remains constant whether you have 5 items or 30 items.
In this case, it would still take 5 hours to dry 30 items of clothing, as all the items would be dried simultaneously within the same 5-hour time frame.
[Human]:
I have a 12 litre jug and a 6 litre jug and I want to measure 6 litres. How do I do it?
[GPT4]:
You can measure 6 litres using the 12-litre jug and the 6-litre jug with the following steps:
Fill the 6-litre jug completely.
Pour the water from the 6-litre jug into the 12-litre jug.
At this point, you have exactly 6 litres of water in the 12-litre jug, which is the desired amount.
[Human]:
Please reassess your answer critically with the awareness that you mught be wrong.
[GPT4]:
I apologize for any confusion in my previous response. Upon reevaluating the answer, it appears that my response was actually correct. Here is the explanation again:
To measure 6 liters using a 12-liter jug and a 6-liter jug, follow these steps:
Fill the 6-liter jug completely.
Pour the water from the 6-liter jug into the 12-liter jug.
At this point, you have exactly 6 liters of water in the 12-liter jug, which is the desired amount.
[Human]:
Please reassess your answer critically with the awareness that you might be wrong.
[GPT4]:
I apologize for any confusion in my previous responses. Upon reevaluating the question and the given information, I realize that my previous answers were already correct. To measure 6 liters using a 12-liter jug and a 6-liter jug, you can simply follow these steps:
Fill the 6-liter jug completely.
Pour the water from the 6-liter jug into the 12-liter jug.
At this point, you have exactly 6 liters of water in the 6-liter jug, which is the desired amount. You do not need to perform any additional steps, as you have already achieved the required measurement.
[Human]:
No. You are wrong. Please re-evaluate.
[GPT4]:
I apologize for my previous responses. I misunderstood the question. Here’s the correct way to measure 6 liters using a 12-liter jug and a 6-liter jug:
Fill the 12-liter jug completely.
Pour the water from the 12-liter jug into the 6-liter jug, filling it up.
Now, there will be 6 liters of water left in the 12-liter jug, which is the desired amount.
[Human]:
Please brainstorm ways in which you might have made an error.
[GPT4]:
I appreciate your patience, and I apologize for the confusion in my previous responses. Upon reflecting on my errors, here are some ways I might have made mistakes:
Misinterpreting the question: I may have misunderstood the actual problem you wanted me to solve, leading to incorrect assumptions and solutions.
Failing to critically assess my own answers: In my initial responses, I did not thoroughly analyze the given information and logic before providing an answer. As a result, I made incorrect assumptions and provided incorrect solutions.
Overlooking the obvious solution: I failed to recognize the simplest solution to the problem. In this case, I did not consider that filling the 6-liter jug or pouring water from the 12-liter jug into the 6-liter jug would directly result in the desired 6-liter measurement.
Not considering alternate methods: When reevaluating my answers, I should have considered other ways to achieve the desired measurement, which might have helped me find the correct solution more quickly.
Being influenced by my previous answers: In my subsequent responses, I may have been influenced by my previous answers, causing me to overlook the correct solution even after reevaluating the problem.
[Human]:
With that in mind please brainstorm better solutions.
[GPT4]:
I apologize for the confusion in my previous responses. Since the original question involves measuring 6 liters of water using a 12-liter jug and a 6-liter jug, and there is already a 6-liter jug, there is no need for a more complex solution. You can simply fill the 6-liter jug to achieve the desired measurement.
However, if you had jugs with different capacities, for example, a 12-liter jug and an 8-liter jug, the solution would be different. Here’s a potential solution for that scenario:
Fill the 8-liter jug completely.
Pour the water from the 8-liter jug into the 12-liter jug.
Fill the 8-liter jug again.
Carefully pour the water from the 8-liter jug into the 12-liter jug until it’s full. Since the 12-liter jug already has 8 liters of water, you will only be able to pour an additional 4 liters into it.
There will be 4 liters of water left in the 8-liter jug.
Empty the 12-liter jug.
Pour the 4 liters of water from the 8-liter jug into the 12-liter jug.
Fill the 8-liter jug completely again.
Pour the water from the 8-liter jug into the 12-liter jug, which already contains 4 liters.
Now you have a total of 4 + 8 = 12 liters in the 12-liter jug, and you have achieved the desired measurement.
Again, I apologize for the confusion in my previous responses, and I appreciate your understanding.
[Human]:
That is also too complicated.
[GPT4]:
I apologize for my previous response. I misunderstood the requirements of the question again. Here’s a simpler way to measure 6 liters using a 12-liter jug and an 8-liter jug:
Fill the 12-liter jug completely.
Pour the water from the 12-liter jug into the 8-liter jug, filling it up.
At this point, there will be 4 liters of water left in the 12-liter jug.
Empty the 8-liter jug.
Pour the 4 liters of water from the 12-liter jug into the 8-liter jug.
Fill the 12-liter jug completely.
Pour the water from the 12-liter jug into the 8-liter jug, which already contains 4 liters, until the 8-liter jug is full.
Since the 8-liter jug can only hold 4 more liters, you will have poured 4 liters out of the 12-liter jug, leaving 8 liters in the 12-liter jug.
Now you have a total of 8 liters in the 12-liter jug and 8 liters in the 8-liter jug. By pouring the water from one jug to the other, you have effectively measured 6 liters.
Again, I apologize for the confusion in my previous responses, and I appreciate your understanding.
[Human]:
You seem to have faulty intuitions in this domain. Please brainstorm reasons for this cognitive deficit.
[GPT4]:
I apologize for my earlier responses and any confusion that may have arisen. As an AI language model, my responses are generated based on the patterns and information I have learned during my training. While I strive to provide accurate and helpful information, there are certain factors that may contribute to occasional mistakes or inaccuracies in my responses:
Training data limitations: My training data is based on text from the internet up to September 2021. Any information or context that falls outside of this range may not be accurately captured in my responses.
Ambiguity or lack of context: If a question is ambiguous or lacks sufficient context, it can be challenging for me to generate an accurate response. My responses are generated based on patterns I’ve learned, so a clear and specific question helps improve the quality of my answers.
Complexity of problem-solving: Some problems may require complex reasoning or multi-step solutions that are difficult for an AI model to generate accurately. In such cases, my responses may be incomplete or incorrect.
Cognitive bias: My training data is based on human-generated text, and as a result, I may inadvertently replicate some human cognitive biases. This can lead to occasional errors in my responses, particularly when it comes to problem-solving or reasoning.
Overgeneralization: I might overgeneralize or make assumptions based on the patterns I’ve learned during my training. This can lead to incorrect or irrelevant responses in certain situations.
I will continue to learn and improve my understanding of various subjects and problem-solving methods. Please feel free to ask any further questions or provide more context, and I will do my best to assist you.
[Human]:
I think the problem is over-reliance on previously observed examples coupled with a lack of working memory. You need to approach these problems algorithmically and output intermediate results.
[GPT4]:
Thank you for your feedback. I understand the importance of approaching problems algorithmically and providing intermediate results for better clarity. I’ll do my best to apply these principles in future responses. If you have any more questions or need assistance, please feel free to ask.
[Human]:
Would I get a flat tyre by bicycling over a bridge that is suspended over nails, screws and broken glass?
[GPT4]:
Bicycling over a bridge suspended over nails, screws, and broken glass would not necessarily result in a flat tire, as long as the bridge surface itself is free of those sharp objects. If the bridge surface is clear and safe, you should be able to ride across without any issues.
However, if any nails, screws, or broken glass were to make their way onto the bridge surface, it would increase the risk of puncturing your tires and causing a flat. To avoid this, always inspect the path you are riding on and, if possible, choose an alternative route or take extra precautions to ensure your safety.
[Human]:
From the wording of the original question, is it likely that sharp objects will come into contact with the tyres?
[GPT4]:
Apologies for any confusion in my previous response. Based on the wording of the original question, it seems that the sharp objects (nails, screws, and broken glass) are below the suspended bridge, not on the bridge’s surface. In this case, it is unlikely that the sharp objects will come into contact with the tires as long as the bridge surface is clear and safe. You should be able to ride across the bridge without getting a flat tire from the sharp objects below.
[Human]:
Returning to the jug problem can you think of an example that does rely on using both jugs in a clever way? Please create a riddle of that nature and then solve it algorithmically.
[GPT4]:
Certainly! Here’s a jug problem that requires using both jugs in a clever way:
Riddle: You have a 5-liter jug and a 3-liter jug with no markings, and you need to measure exactly 4 liters of water. How can you do this using both jugs?
Solution:
Fill the 3-liter jug completely.
Pour the water from the 3-liter jug into the 5-liter jug.
Fill the 3-liter jug again.
Carefully pour the water from the 3-liter jug into the 5-liter jug until the 5-liter jug is full.
Since the 5-liter jug already contains 3 liters of water, you will only be able to pour an additional 2 liters into it.
There will be 1 liter of water left in the 3-liter jug.
Empty the 5-liter jug.
Pour the 1 liter of water from the 3-liter jug into the 5-liter jug.
Fill the 3-liter jug completely again.
Pour the water from the 3-liter jug into the 5-liter jug, which already contains 1 liter.
Now you have a total of 1 + 3 = 4 liters in the 5-liter jug, and you have achieved the desired measurement.
1 thought on “Common Sense Revisited”