Eleven Techniques, Zero Winners
Prompt Engineering for LLMs, Measured Against a Sentence That Instructs Nothing
Impossible d'ajouter des articles
Désolé, nous ne sommes pas en mesure d'ajouter l'article car votre panier est déjà plein.
Veuillez réessayer plus tard
Veuillez réessayer plus tard
Échec de l’élimination de la liste d'envies.
Veuillez réessayer plus tard
Impossible de suivre le podcast
Impossible de ne plus suivre le podcast
Bénéficiez gratuitement de Standard pendant 30 jours
5,99 €/mois après la période d’essai. Annulation possible à tout moment
Acheter pour 13,99 €
-
Lu par :
-
AI Voice A synthetic voice
-
De :
-
Nikolaos Broikos
Ce titre utilise une narration à voix virtuelle
La voix virtuelle est une narration générée par ordinateur pour les livres audio.
Not one of them beat it.
Think step by step. Take a deep breath. You are an expert. Please. I will tip you. Every one of them arrived in working practice as something everybody knew, and not one arrived with a number attached. So they were measured properly: the same tasks, the same graders, three repeats each, every call isolated, and a control designed to look exactly like a technique while instructing nothing.
Eleven techniques went in. None came out ahead on accuracy. Seven of them significantly damaged the shape of the answer, which is the part your code depends on. The single most recommended instruction in the field took responses that were clean every single time down to fewer than one in three.
This is not an opinion piece about AI hype. It is a measurement programme with the trial data published, the graders published, and the gates that had to pass before any of it was allowed to run. Where the author's own instruments turned out to be wrong, that is in the book too, with the numbers it changed.
If you have ever added a phrase to a prompt because somebody confident said it worked, this is the book that tells you what it cost you.
adbl_web_anon_alc_button_suppression_t1
Aucun commentaire pour le moment