This is an excellent analysis, and it also highlights the importance of 'writing for the llms'. I know we technically call this data poisoning, but its a good illustration of how one *influences the AI* especially if you have an idea outside of consensus.
You hit on something that I think is critical, and frequently missed, when discussing LLMs and their use, whether by government officials or anyone else. Knowing *how* to ask good questions, and what makes a question good, is as important as knowing that you can just ask questions. And I think this is the type of skill that one only acquires after using LLMs for a long time, surely more than a dozen hours. It takes a while to learn how to extract value from LLMs. And in an administration as chaotic as Trump's appears to be, taking the time to learn how to use an LLM properly seems unlikely.
I've always found it fascinating how old books and old librarians play such a prominent role in sci fi. While inefficient and dusty and too often caught up in weird emotions, books have obvious utility as a stable information technology.
I don't think it was LLM. It was our president, who isn't any smarter than that. His "brilliance" was showing in the previous term quite well, long before LLM got invented.
Do you not realize that discovering the source to be Navarro means that LLMs were almost certainly *not* used to determine the tariff levels? Navarro is a Trump Administration trade advisor, the key official behind the policy. The formula evidently came directly from him, not via LLMs. So this was not actually an instance of vibe governing. Right?
Hmm, I can't actually find confirmation that Navarro ever explicitly advocated tariffs proportional to trade balances prior to Apr 2. If he hadn't, that calls into question whether the LLM suggestion can really be attributed to him (and also whether he was the direct source of the Trump policy formula).
A (Navarro is Trump's trade czar) -> Z (Navarro's tariff formula became Trump policy)
Simultaneously:
A (Navarro is Trump's trade czar) -> B (Trump Administration uses language like "on even-playing fields when it comes to trade deficit") -> C (You and others ask LLMs for tariffs calculated for that goal) -> D (The LLMs, associating that language with Navarro, give you Navarro's tariff formula)
The fact that LLMs give you Navarro's tariff formula when prompted with Navarro's language (language other trade economists generally do not use) tells you nothing about whether the Trump Administration consulted LLMs to determine the tariff formula. And why would Navarro ask an LLM for what tariff formula to use? He has apparently known what tariff formula he favors since at least 2011.
This is an excellent analysis, and it also highlights the importance of 'writing for the llms'. I know we technically call this data poisoning, but its a good illustration of how one *influences the AI* especially if you have an idea outside of consensus.
You hit on something that I think is critical, and frequently missed, when discussing LLMs and their use, whether by government officials or anyone else. Knowing *how* to ask good questions, and what makes a question good, is as important as knowing that you can just ask questions. And I think this is the type of skill that one only acquires after using LLMs for a long time, surely more than a dozen hours. It takes a while to learn how to extract value from LLMs. And in an administration as chaotic as Trump's appears to be, taking the time to learn how to use an LLM properly seems unlikely.
I've always found it fascinating how old books and old librarians play such a prominent role in sci fi. While inefficient and dusty and too often caught up in weird emotions, books have obvious utility as a stable information technology.
Trantor and Asimov imprinted this on me!
I think we're deep inside a mash-up - imagine what a genre-bending, glitchy musical-comedy-thriller directed by a confused algorithm might look like. The fun has truly begun: https://rajeshachanta.substack.com/p/don-quixote-and-his-dreamcoat
I don't think it was LLM. It was our president, who isn't any smarter than that. His "brilliance" was showing in the previous term quite well, long before LLM got invented.
Do you not realize that discovering the source to be Navarro means that LLMs were almost certainly *not* used to determine the tariff levels? Navarro is a Trump Administration trade advisor, the key official behind the policy. The formula evidently came directly from him, not via LLMs. So this was not actually an instance of vibe governing. Right?
I think there is usually a citation somewhere but considering the answers I'm pretty sure this qualifies.
Hmm, I can't actually find confirmation that Navarro ever explicitly advocated tariffs proportional to trade balances prior to Apr 2. If he hadn't, that calls into question whether the LLM suggestion can really be attributed to him (and also whether he was the direct source of the Trump policy formula).
Almost certainly the true causal graph here was:
A (Navarro is Trump's trade czar) -> Z (Navarro's tariff formula became Trump policy)
Simultaneously:
A (Navarro is Trump's trade czar) -> B (Trump Administration uses language like "on even-playing fields when it comes to trade deficit") -> C (You and others ask LLMs for tariffs calculated for that goal) -> D (The LLMs, associating that language with Navarro, give you Navarro's tariff formula)
The fact that LLMs give you Navarro's tariff formula when prompted with Navarro's language (language other trade economists generally do not use) tells you nothing about whether the Trump Administration consulted LLMs to determine the tariff formula. And why would Navarro ask an LLM for what tariff formula to use? He has apparently known what tariff formula he favors since at least 2011.