Zum Hauptinhalt springen

Gastvortrag (30.05.2024; 10-12 Uhr) - Anna-Maria De Cesare (TU Dresden)

„Beyond the surface: Evaluating the quality of ‘texts’ generated in Italian by Large Language Models“

automatically vs. human generated texts
Foto : Picture automatically generated by Art AI

Alle Interessierten sind sehr herzlich zum englischsprachigen Gastvortrag von Prof. Dr. Anna-Maria De Cesare (Technische Universität Dresden) Beyond the surface: Evaluating the quality of ‘texts’ generated in Italian by Large Language Models eingeladen.

*Zeit: Donnerstag, 30. Mai 2024, 10:15-11:45 Uhr*

*Ort: Universität Potsdam, Campus ‚Am Neuen Palais‘, Raum 1.19.4.15*

Den Flyer zur Veranstaltung finden Sie hier.

 

Beyond the surface: Evaluating the quality of ‘texts’ generated in Italian by Large Language Models

Anna-Maria De Cesare (Technische Universität Dresden)

Abstract:

The advent of large language models (LLMs) has revolutionized our ability to generate linguistic output in a wide range of languages. According to many evaluations, LLM-generated outputs are coherent, high-quality texts that are in many ways comparable to and indistinguishable from human-written texts. The aim of this talk is to reflect on how we can verify these claims and assess whether LLM-generated output is indeed human-like. Based on the results of several case studies analyzing data drawn from a self-assembled corpus of GPT-4-generated and human-written biographies, I will show that LLM-generated texts appear well-written only at first glance. On closer inspection of textual parameters (related to the referential dimension of textual organization), these texts tend to be repetitive and monotonous.

Veröffentlicht

Online-Redaktion

Marta Lupica Spagnolo