LCB-Bench v1.1: Benchmarking Large Language Models for Cognitive Bias Detection

20260 citationsPreprintgreen Open Access

Authors

Abstract

LCB (LLM Cognitive Bias Benchmark) v1.1: Updated evaluation results with complete GPT-4o coverage (1500/1500 cases scored, 0 errors). 1,500 human-authored test cases measuring 30 cognitive biases across 7 categories. Includes evaluation harness and results for 4 frontier models (GPT-4o, GPT-4o-mini, Gemini 2.5 Flash, Claude Sonnet 4.6). LCB Scores range from 69.0 to 83.6/100. Related to DOI 10.5281/zenodo.19185736 (v1.0).

Topics & Keywords

Artificial Intelligence in Healthcare and Education Neurobiology of Language and Bilingualism Machine Learning in Healthcare

UN Sustainable Development Goals

Quality Education

Publication Details

Published in: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.19189848