Recent searches
Search options
Oh. My. Fucking. God.
They got the tariff formula by asking some AI models.
Four out of four models agreed that this half assed 2nd grade arithmetic formula was a great approach to geopolitics.
"This is vibe governing" b/w this is the first demonstrable AI safety issue.
https://x.com/krishnanrohit/status/1907587352157106292
I mean who could have predicted such a grossly stupid use of AI to make critical decisions?
Well, I guess we all did.
As much as we might like to believe this, it is so improbable as to be impossible.
First off, there’s straight up no way all 4 models give the “same” output: it’s astronomically unlikely to happen once, much less when 2 different people (some DOGE fool and this person) ask the same question. So it could be similar, but that’s it.
Secondly generative algorithms are non-deterministic. Give the exact same model the exact same inputs, and you will NOT get the exact same answer twice. It will give you different variations, perhaps with lots of similarities, each time you ask. Try it.
So even if this IS the answer they got from the stochastic text extruder today, there is zero reason to believe that the answer they received today is the same that some DOGE flunky got a few days ago.
This feels either dumb or rage bait. Note how little of the prompts or the answers are actually present in the screenshots. Maybe they just asked a really vague, simple question and got similar results.
I hate them (and AI) as much as anybody. But this isn’t credible at face value.
@paco @Steve i've seen a bunch of people independently verify that at least ChatGPT gives this answer when asked a similar question.
Presumably that answer is given because, in a sense, it does make sense as a way to "even the playing field" though I agree it's not necessarily true that they actually got the answer from an LLM.
That being said, it's also true that the list of countries they want to tariff includes several antarctic islands that are completely uninhabited.... so there's a very strong "last minute rushing to complete a term paper due tomorrow" vibe here. No one spent any real amount of time coming up with these numbers, or they would have noticed the uninhabited islands.
https://www.axios.com/2025/04/02/trump-tariffs-australia-uninhabited-heard-mcdonald-islands
I would agree viz repeatable output variation. It's highly unlikely - if not impossible - to achieve the same exact output given a simple prompt across models. However, I was referring to the document published by USTR, which certainly has all the grammatical and vocabulary hallmarks of genAI
How am I meant to know there was no-one living on the archipelago I slapped with tariffs?
So what if I am an AI?
So here's what they have published. It reads like it was written by GenAI and mostly gibberish
https://ustr.gov/issue-areas/reciprocal-tariff-calculations
@cryptadamist I guess the AI doomers were half right. It just turns out that the problem isn't the super-intelligence of the AI, it's the super-stupidity of the operators.