Note that the prompt given here does not ask for a way of calculating the effective foreign tariffs imposed on the US (what the formula is used to dishonestly present) but to calculate the ideal US tariffs to level the trade deficit.
That is, the LLMs do not give the formula the US actually used for setting US tariffs with this prompt asking for ideal US tariffs, instead, the response to this prompt about ideal US tariffs gives the calculation that the Administration used to calculate the effective foreign tariffs, not the US tariffs.
Also, given that tariff rates must be a percentage, any answer is probably going to end up being a ratio, and the prompt naturally indicates that the numerator should be the trade deficit the US has with the country. So. this is essentially expressing surprise that when suggested half of the formula in the prompt, the models all come up with the same (and I would argue quite obvious once you've gotten that far) denominator.
I don't think this indicates that that used an LLM, I think it indicates that they decided that all trade deficits shoule be ascribed to manipulation that was effectively tariff-like, and presented as a tariff rate imposed by the foreign country, and, having made that decision, the formula fell out naturally, and is what any reasonably knowledgeable human given the assignment—as well as any non-hallucinating, modestly capable LLM—would produce given the way the assignment already implicitly provides both the shape and one of two key pieces of the formula.
(Of course, expressing surprise at AI doing something in a response and not realizing it was mostly just regurgitating what was already decided by the prompter and fed to it in the prompt in a slightly different form is quite common is quite common these days, though using that mistake as a basis for an accusation of using AI is somewhat novel.)
That is, the LLMs do not give the formula the US actually used for setting US tariffs with this prompt asking for ideal US tariffs, instead, the response to this prompt about ideal US tariffs gives the calculation that the Administration used to calculate the effective foreign tariffs, not the US tariffs.
Also, given that tariff rates must be a percentage, any answer is probably going to end up being a ratio, and the prompt naturally indicates that the numerator should be the trade deficit the US has with the country. So. this is essentially expressing surprise that when suggested half of the formula in the prompt, the models all come up with the same (and I would argue quite obvious once you've gotten that far) denominator.
I don't think this indicates that that used an LLM, I think it indicates that they decided that all trade deficits shoule be ascribed to manipulation that was effectively tariff-like, and presented as a tariff rate imposed by the foreign country, and, having made that decision, the formula fell out naturally, and is what any reasonably knowledgeable human given the assignment—as well as any non-hallucinating, modestly capable LLM—would produce given the way the assignment already implicitly provides both the shape and one of two key pieces of the formula.
(Of course, expressing surprise at AI doing something in a response and not realizing it was mostly just regurgitating what was already decided by the prompter and fed to it in the prompt in a slightly different form is quite common is quite common these days, though using that mistake as a basis for an accusation of using AI is somewhat novel.)