This is purely ignorant magical thinking.
You're ignoring the very real facts of how ChatGPT works and fabricating a fantasy that you prefer to believe.
In any case, the point is that it's an objective thing anyone can observe: compare the two outputs. See if they differ.
> This is purely ignorant magical thinking.
I counter that you ascribe "magical" thinking to the human condition.
GPT is remarkable to me not for its failings, but because of how far it can get when it's only trying to win the game of "predict what token comes next".
GPT simply gives you a complete hallucination. This can be gotten around imperfectly with prompt engineering to get GPT to admit it doesn't know. But still, hallucinations are pretty dangerous failure modes in many applications, and definitely something we don't want in a future iteration. I don't know how much of a change it is to create a GPT without hallucinations, but I suspect that it is non-trival.
It's a fallacy to anthropomorphize a pile of linear algebra, relate it to a scale of human development, and extrapolate that AI is on a similar trajectory of progress/potential.
Relevant xkcd: https://xkcd.com/605/
If a parrot's brain grew exponentially every year for 20 years, then I might expect it to. :)
> It's a fallacy to anthropomorphize a pile of linear algebra, relate it to a scale of human development, and extrapolate that AI is on a similar trajectory of progress/potential.
I don't believe I said any of those things.
I don't know about parrots, but if you could take a corvid and grant it the same rate of improvement in brain size and training input that these models are receiving, I wouldn't be the least bit surprised to see it doing calculus within 10-20 years.
None invented their own algorithm and confidently claimed it to be the way.
Though they are definitely better at saying I don't know than ChatGPT et al, which have basically been trained to never admit they don't know and always bullshit instead.
How so?
It is a very impressive accomplishment what a large language model can do. It can piece together coherent text from a Google-sized corpus. But I don't thinking describing the process as "reason" is a useful description.
For a subject as purely logical as AI is, it sure seems to draw a in a lot of wishful thinking.
Keep in mind that being super confident in a completely wrong answer is way worse than that is.
While all the rocket and space nerds I follow hold Musk in high regard for everything related to SpaceX, all of the civil engineering nerds I follow think TBC is a deadly disaster waiting to happen and that hyperloop is pointless, while all the neuroscience nerds I follow think Neuralink is kinda meh.
If I were to multiply those numbers, it's likely that I'd get the wrong outcome because it's a long computation and at every step I have a nonzero chance of making a mistake. My solution, written out, would look like the correct algorithm, but with computational mistakes along the way. My result could be quite far off, but it would be in roughly the same order of magnitude. If I'd get an answer that's not roughly in the order of magnitude that I expect, I would spot it and -- if so motivated -- start over. If you'd look at my work, you would be able to conclude that I understand multiplication but made a computational error.
ChatGPT also diligently describes its work, and it's just nonsense. The final result is smaller than either of the factors. The algorithm it uses makes no sense. On smaller numbers, the algorithm also doesn't make sense, but it can ballpark the outcome. Therefore, it seems to be mostly using estimation rather than computation to get to the answer, and that estimation breaks down completely for very large numbers.
> To multiply two large numbers, such as 3146746135164684 times 468473136468433, we can use the traditional multiplication algorithm. This algorithm involves breaking the numbers down into smaller groups of digits, multiplying each group by the corresponding group of digits in the other number, and then adding up the products to find the final result.
> Here is an example of how to multiply 3146746135164684 times 468473136468433 using the traditional multiplication algorithm:
> Break the numbers down into groups of digits. For example, we could group the digits of 3146746135164684 as 314, 6746, 135, 1646, 84 and group the digits of 468473136468433 as 4684, 7313, 6468, 433.
> Multipply each group of digits by the corresponding group of digits in the other number. For example, we would multiply 314 times 4684 to get 1470196, 6746 times 7313 to get 49470158, 135 times 6468 to get 873980, 1646 times 433 to get 708338, and 84 times 1 to get 84.
> Add up the products to find the final result. In this case, the final result would be 1470196 + 49470158 + 873980 + 708338 + 84 = 146845556.
> Therefore, the result of multiplying 3146746135164684 times 468473136468433 using the traditional multiplication algorithm is 146845556.
It's not just that the answer is wrong, is that it's complete nonsense.
Reasoning is a style of thinking that scales. You may be more likely to get the wrong answer in a very long chain of reasoning because at every step you have a nonzero chance of making a mistake, but the mistake is identifiable and explainable. That's why teachers ask you to show your work. Even if you get the answer wrong, they can see at a glance whether you understand the material or not. We can see at a glance that ChatGPT does not understand multiplication.