Abstract
As Large Language Models (LLMs) increasingly integrate with external knowledge sources like the internet, the ability to discern information becomes crucial. This research investigates whether these models adequately weigh information—updating more for reliable sources (source discernment) and more when claims bring priors closer to the truth (truth discernment). We formalize this as information discernment and introduce Learn2Discern (L2D), an experimental framework and benchmark grounded in three normative axioms with interpretable metrics.
To establish external validity, a pre-registered user study (n=299) confirms that real LLM users endorse all three axioms, reporting that violations reduce their trust and usage intent. Across 13 models and nearly 670K trials, we observe consistent failures in both dimensions: models perform near chance on source and truth discernment, relying on source popularity twice as much as reliability, and update roughly equally whether a claim improves or worsens their position relative to the ground truth.
Models integrate external knowledge most effectively on datasets where their priors are already the most accurate. Newer and larger models improve truth discernment but not source discernment, indicating a blind spot that model complexity does not address. We identify simple inference-time interventions that improve both forms of discernment. We release our dataset and survey as a testbed for a core alignment property that scales in importance as LLMs replace traditional search.
Blogger's Review: This study provides a deep dive into the limitations of large language models in information processing, particularly in source and truth discernment. Future improvements should not only focus on scaling model size but also on accurately assessing information sources to enhance user trust and intent to use.