Text Understanding in GPT-4 vs Humans

TL;DR

This study compares GPT-4's text comprehension abilities with humans using standardized tests, revealing GPT-4's comparable or superior performance, especially on difficult passages, and signs of genuine understanding like inference and generalization.

Contribution

It provides a detailed comparison of GPT-4 and human comprehension, demonstrating GPT-4's ability to understand and generalize on complex texts better than high school and university students.

Findings

GPT-4 performs slightly better than humans on standard comprehension tests.

GPT-4 significantly outperforms students on difficult passages.

GPT-4 shows signs of genuine understanding through inference and generalization.

Abstract

We examine whether a leading AI system GPT4 understands text as well as humans do, first using a well-established standardized test of discourse comprehension. On this test, GPT4 performs slightly, but not statistically significantly, better than humans given the very high level of human performance. Both GPT4 and humans make correct inferences about information that is not explicitly stated in the text, a critical test of understanding. Next, we use more difficult passages to determine whether that could allow larger differences between GPT4 and humans. GPT4 does considerably better on this more difficult text than do the high school and university students for whom these the text passages are designed, as admission tests of student reading comprehension. Deeper exploration of GPT4 performance on material from one of these admission tests reveals generally accepted signatures of genuine understanding, namely generalization and inference.