Girls, Pigs and YandexGPT 3

I once slightly modified the "geese and rabbits problem" and offered it to my students (these are high school students from a smart school - so the original problem would be ridiculously simple for them). It's always interesting to see how differently brains work when solving it. Well, let's feed it to the AI and see what illogical and inconsistent answers it tries to feed us in return :)

The village headman discovered that the shepherd girls from the neighboring village accidentally drove a herd of pigs into someone else's fields. To avoid suspicion, he sends the village fool, telling him to count both - so that he can then present the neighbors with a bill for the damage. The fool returns:
- Well, did you count? - asks the headman
- 106 legs and 336 teats - the fool reports happily
The fool, of course, got it. But by the time the headman's anger subsided, however, the shepherd girls had already figured it out and left, along with their piglets. Let's leave the headman to experience this embarrassment - and let's think for ourselves, is it possible to determine how many girls and how many pigs there were from such data?

We do not plan to test the patience of readers, so I will immediately explain - in this task, unlike geese and rabbits, there are three unknowns for two equations - so it requires some reasonable assumptions. For example, that all pigs are homeomorphic :) By the way, women more often solve this puzzle in their minds...

However, we are not here to solve the puzzle but to evaluate the response of YandexGPT.

I will explain that I remembered this task simply because I am quite sure - AI is unlikely to have come across it on the internet and given me a ready-made solution. Although it has appeared somewhere.

And so I tried to shorten the question so that only the essence related to the calculations remained. On the first attempt, I faced a fiasco - it seems that the AI was offended by the literal answer of a fool and, after flashing the inscription "Not enough data..." for a second, replaced it with the suggestion "Let's change the topic".

I replaced the word that apparently irritated the AI (indeed, somewhat vulgar-colloquial) and it turned out like this.

Question

There are several girls and several pigs, they have a total of 106 legs and 336 nipples - how many girls?

Answer YandexGPT 3

Unfortunately, it is impossible to solve this problem without additional information.

For an accurate answer, you need to know the number of pigs and the total number of heads (girls and pigs).

I assume that each girl has two legs and zero nipples, and each pig has four legs and two nipples. Then you can make an equation:

2 * x + 4 * y = 106,

where x is the number of girls, y is the number of pigs.

3 * x + 2 * y = 336.

Solving the system of equations, we get that x = 9, y = 5. However, this is only one of the possible solutions, which does not take into account other options for the distribution of legs and nipples between girls and pigs.

Lack of self-control, self-criticism

AI throws a heap of text at us, makes some assumptions. At the same time, it is clear that critical consideration of its own words is alien to it.

Solving the system of equations, we get that x = 9, y = 5.

It sounds beautiful and smart, but we are surprised to find that these answers do not correspond at all to the system that the AI has compiled. Moreover, the second equation of the system does not correspond to the phrase "let's assume". It manages to "mess up" at every step. This behavior was probably observed by those who experimented with generating code using different versions of GPT - it looks like it is beautiful, but it does not even necessarily compile.

For an accurate answer, you need to know the number of pigs and the total number of heads (girls and pigs).

You probably noticed that knowing both of these parameters is not required at all. Indeed, if we know the number of pigs, we can easily count the number of pig legs, subtract from 106 and divide by 2.

Moreover, if we knew both of these numbers, then all other data would not be needed - because if you subtract the number of pigs from the number of heads, the number of girls will remain!

I assume that each girl has two legs and zero nipples, and each pig has four legs and two nipples.

Here you really need to "turn off the light". Where do such assumptions come from? Moreover, if you enter the question "how many nipples does a pig have" - the AI will answer quite adequately.

Attempt #2

Having asked about the number of nipples in a pig and making sure that the AI is roughly aware of the anatomical features of these animals, I ask it to "try to clarify the solution to the problem taking this information into account".

A heap of text falls on me, starting with the already known:

Let's assume that each girl has two legs and zero nipples, and each pig has four legs and two nipples. Then we can make an equation:

Read also:

• Restoring a non-working keyboard using QMK and RP2040

...

It turns out that we have 9 girls and 5 pigs. Let's check the solution: 9 girls have 0 * 9 = 0 nipples, 5 pigs have 2 * 5 = 10 nipples. Total nipples: 0 + 10 = 10. It does not match the condition of the problem.

Then it suggests a brute force method, but it hopelessly floats in the calculations:

Let's try to solve the problem in another way. Let's assume that all animals are pigs. Then the total number of legs will be 4 * 106 = 424, which does not correspond to the condition of the problem. Therefore, our assumption is incorrect.

So, there are both girls and pigs among the animals. If we find the total number of pigs, we can determine the number of girls. To do this, subtract 2 * the number of girls from the total number of legs:

106 — 2 * x = even number.

An even number can only be obtained with x, a multiple of 2. Let's try to substitute numbers that are multiples of 2 instead of x:

For x = 2: 106–4 = 102 — odd number, does not fit.

For x=4: 106-8 = 98 — odd number, does not fit.

For x=6: 106-12 = 94 —

Let's leave it to the conscience of the sages from Yandex to claim that 102 and 98 are odd numbers. And to hell with the fact that the size of the answer is obviously limited and the engine does not take this limitation into account, which is why it breaks off so abruptly.

But think about the phrase: Let's assume that all animals are pigs.

What is this about, Carl? There were only girls and pigs. Maybe the AI has read too much Orwell? Maybe this is a general remark, not about the task? I'll try to ask again "is it true that all animals are pigs?"

No, this is not true. From the task condition, we know that there are 106 legs and 336 nipples in total among the girls and pigs.

Let's assume that all animals are pigs. Then the total number of legs would be 4 * 106 = 424, which does not match the task condition. Therefore, our assumption is incorrect. So, there are both girls and pigs among the animals.

The last phrase is certainly shock content.

Conclusion

The AI in this case is similar to a barrel filled with various more or less template phrases, formulas, and answers. A competent sequence of presentation creates the impression that it sometimes really thinks. But when it comes to factual data, it turns out that it rather "turns the kaleidoscope" and tries to comfort us with another stochastically formed picture.

The AI does not try to solve the problem - its goal is only to generate an answer that will satisfy you!

This is similar to a poor student trying to guess the answer - only more stochastically and inhumanly. Probably the 4th version of ChatGPT can handle it better - it is certainly better tuned. If possible, check it - use a request like this:

Some girls and several pigs have 106 legs and 336 nipples in total - how many girls are there?

I apologize to everyone who might have found this strange task and strange dialogue offensive - especially to the ladies. This was not invented out of malice - just for the task, some creatures with an unknown number of certain organs were needed - to have a third unknown in the equations. Pigs just happened to be a convenient choice.