Sebastian Hellmann

Marcel Binz, Elif Akata, Abdullah Almaatouq et al.

Large language models (LLMs) are increasingly used as surrogates for human participants, but it remains unclear which models best capture human behavior and why. To address this, we introduce Psych-201, a novel dataset that enables us to measure behavioral alignment at scale. We find that post-training -- the stage that turns base models into useful assistants -- consistently reduces alignment with human behavior across model families, sizes, and objectives. Moreover, this misalignment widens in newer model generations even as base models continue to improve. Finally, we find that persona-induction -- a popular technique for eliciting human-like behavior by conditioning models on participant-specific information -- does not improve predictions at the level of individuals. Taken together, our results suggest that the very processes that are currently employed to turn LLMs into useful assistants also make them less accurate models of human behavior.

Milan Dojchinovski, Julio Hernandez, Markus Ackermann et al.

In the past decade, the DBpedia community has put significant amount of effort on developing technical infrastructure and methods for efficient extraction of structured information from Wikipedia. These efforts have been primarily focused on harvesting, refinement and publishing semi-structured information found in Wikipedia articles, such as information from infoboxes, categorization information, images, wikilinks and citations. Nevertheless, still vast amount of valuable information is contained in the unstructured Wikipedia article texts. In this paper, we present DBpedia NIF - a large-scale and multilingual knowledge extraction corpus. The aim of the dataset is two-fold: to dramatically broaden and deepen the amount of structured information in DBpedia, and to provide large-scale and multilingual language resource for development of various NLP and IR task. The dataset provides the content of all articles for 128 Wikipedia languages. We describe the dataset creation process and the NLP Interchange Format (NIF) used to model the content, links and the structure the information of the Wikipedia articles. The dataset has been further enriched with about 25% more links and selected partitions published as Linked Data. Finally, we describe the maintenance and sustainability plans, and selected use cases of the dataset from the TextExt knowledge extraction challenge.