A Comprehensive Evaluation of Cognitive Biases in LLMs
This work addresses the issue of cognitive biases in LLMs for AI safety and fairness researchers, providing a comprehensive evaluation tool, but it is incremental as it builds on prior findings.
The researchers tackled the problem of evaluating cognitive biases in large language models by developing a novel test framework and benchmark dataset, resulting in evidence of all 30 tested biases across 20 models, with a dataset of 30,000 tests.
We present a large-scale evaluation of 30 cognitive biases in 20 state-of-the-art large language models (LLMs) under various decision-making scenarios. Our contributions include a novel general-purpose test framework for reliable and large-scale generation of tests for LLMs, a benchmark dataset with 30,000 tests for detecting cognitive biases in LLMs, and a comprehensive assessment of the biases found in the 20 evaluated LLMs. Our work confirms and broadens previous findings suggesting the presence of cognitive biases in LLMs by reporting evidence of all 30 tested biases in at least some of the 20 LLMs. We publish our framework code to encourage future research on biases in LLMs: https://github.com/simonmalberg/cognitive-biases-in-llms