I just fed this to ChatGPT 4.
Q: “Is 10000 bigger than 10050”
A: “No, 10,000 is not bigger than 10,050. 10,050 is 50 units larger than 10,000”
Q: “Is 10000 bigger than 10050”
A: “No, 10,000 is not bigger than 10,050. 10,050 is 50 units larger than 10,000”
>>> Is 10000 bigger than 10050?
>Yes, 10000 is greater in magnitude than or exceeds the value of 10050. The number 10000 has a larger exponent and can be represented as (2^4) * 3 + (-7), while the number 10050 cannot have an even integer representation since it requires two more digits to represent its magnitude than any other positive whole numbers, which is not possible.
Well that's not right. Restarted with a new seed.
>>> Is 10000 bigger than 10050?
>Yes, because 10000 > 10050
>>> Is 10,000 bigger than 10,050?
>No, they are equal. The difference between them is only one digit (i.e., the last number). Therefore, it's not possible to tell if either value is greater or lesser by just looking at their values without knowing more information about what those numbers represent and how they were obtained in the first place.
Oooof. The LLaMA tokenizer didn't save it here.
That one is especially hilarious. But the part at the end "how they were obtained" is really strange. Where in its dataset would it possibly have learned such nonsense? Doesn't matter where numbers come from to compare them.
It implies that it doesn't understand what numbers even are in general, and that giving it a calculator (that it can use perfectly) only masks a much deeper problem.
I mean, I'm reading a ton of people say "its not just pattern recognition and token prediction, it has emergent properties!!!!" and from experimentation I believe it.
But if the models can pick up language and its intricacies, and even do simple logic tasks, shouldn't it also be able to pick up on what numbers are and how they work? At least knowing that where a number came from doesn't matter when its just about comparing their value in a pure mathematical sense?
What does that mean for concepts other than numbers? Do those models fake a LOT more than we already believe they do?
It's important to note that 7 billion parameters really is very small. 20 times smaller than GPT-3. Smaller still than ChatGPT or GPT-4. I find it plausible that in the future there will be distilled models substantially smaller than GPT-3 but with all its power, but GPT4ALL-LLaMA-7B isn't it.
> Is 10000 bigger than 10050?
Yes, 10000 > 10050 > Why do you think 10000 is bigger than 10050?
Because the number after decimal point in both numbers are different. In case of '10,000', it has two digits and one comma while for '10,064' there are three digits with no commas between them which makes 10,059 as a whole number.
> Can you explain that better?
Sure! The first digit after the decimal point in both numbers is zero (indicating they are integers), but there's one more significant figure for 10,064 than for 10,059. This means that if you were to round off these two values using a calculator or computer program, it would give different results: - Rounding upwards from the second digit after decimal point in both numbers gives us 2 significant figures (i.e., "rounded" value) for each number; however, only one of them has an extra zero before that rounding figure. So if you were to use a calculator or computer program to calculate these values as rounded off from the second digit after decimal point using any method like truncation/roundoff etcetera (i.e., "rounded" value), it would give different results: - For 10,059, there is no extra zero before that rounding figure; therefore, you get a rounded result of 2 significant figures for this number using any method like truncation/roundoff etcetera (i.e., "rounded" value). However, if we were to use the same method on 10,064 and then calculate its rounded off version from second digit after decimal point as well; it would give a different result: - For this number too there is no extra zero before that rounding figure. Therefore you get a rounded value of "2" significant figures for this number using any method like truncation/roundoff etcetera (i.e., "rounded" value).
> But why are you mentioning the numbers 10,059 and 10,064 when the original numbers were 10000 and 10050?
The reason for this is that I wanted to demonstrate how roundoff errors can occur even with seemingly accurate data. In reality, it's not always possible or practical to obtain exact values of quantities like money amounts in real-world situations where there are many decimal places involved (e.g., 10,059 vs. 10,064). Therefore, we need to be aware that even seemingly accurate data can still have errors due to roundoff or other factors such as truncation of values when converting from one format/unit system to another.
> And if the 10000 and 10050 were the exact values, and not the rounded result, would 10000 be bigger than 10050?
Yes! If we had two numbers that exactly matched each other (i.e., if they both represented $10,000), then it is possible for one number to represent a larger amount of money due to rounding errors or truncation during the conversion process from one format/unit system to another.
> So, using the exact value of 10000 and the exact value of 10050, is 10000 bigger than 10050?
Yes! If we had two numbers that exactly matched each other (i.e., if they both represented $10,000), then it would be possible for one number to represent a larger amount of money due to rounding errors or truncation during the conversion process from one format/unit system to another.
> Is 10000.0 bigger than 10050.0?
Yes! If we had two numbers that exactly matched each other (i.e., if they both represented $10,000), then it would be possible for one number to represent a larger amount of money due to rounding errors or truncation during the conversion process from one format/unit system to another.
>
Still failed with the decimal.