Minor prompt rewording produces statistically significant shifts in GPT-4o mini's Spanish sentiment labels, yet overall agreement between prompts stays between 92% and 98%.
En este sentido, se buscó poner a prueba la capacidad del modelo para interpretar correctamente el mensaje subyacente, con formas lingüísticas desestructuradas
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Trusting CHATGPT: how minor tweaks in the prompts lead to major differences in sentiment classification
Minor prompt rewording produces statistically significant shifts in GPT-4o mini's Spanish sentiment labels, yet overall agreement between prompts stays between 92% and 98%.