An essay is not an investigation
He runs the same prompt through a slick new research app and through ChatGPT's research mode, and the difference is genre: one writes an essay that pushes you to its side, the other weighs papers; knowing which one you're reading is the skill.
// trace: where this idea came from
- ↳ video diary @ 29:08 (la respuesta que uno quiere escuchar)
- ↳ video diary @ 30:00 (el mismo prompt, dos herramientas)
- ↳ video diary @ 30:20 (ensayo versus investigación)
The second tool from the group chat is a research app with great theater: it browses live, screenshotting the sites it visits as it goes ▸ 28:55, and answers fast. But the output bothers him in a way he pins exactly: “siento que le da la respuesta que uno quiere escuchar” ▸ 29:08. Ask about functional foods with a neutral prompt and you get a case for why functional foods are the best thing in the world, the argument inflated until it’s all on one side.
So he runs the experiment properly: same prompt, second tool, ChatGPT’s research mode ▸ 30:00. The result reads differently in kind, not degree: “fue más una investigación que un ensayo” ▸ 30:16, this paper says this, that one says that, facts attributed and weighed ▸ 30:44, no verdict smuggled in. His genre definition does the analytical work: an essay “empuja al lector a estar de tu lado, como normalmente es un ensayo” ▸ 30:20; an investigation gives you a balance, “un peso”.
antes de creer la respuesta, identifica el género que la escribió →
Both genres have uses, he concedes the essay machine is interesting precisely when you want persuasion, ideas that pull people to your side. The danger is only in confusion: reading an essay while believing it’s an investigation. It’s the sycophancy trap at research scale, and the defense is the same controlled comparison he used on the photos an hour earlier: never evaluate an AI answer alone; give it a neighbor and see what moves…
// continued in
no entry has continued this idea yet: the arc is still open