Thinking Fast and Slow in Large Language Models
This research addresses the problem of evaluating emergent abilities in LLMs for AI safety and human-AI interaction, though it is incremental as it applies existing psychological methods to new models.
The study investigated whether large language models (LLMs) exhibit human-like intuitive errors, finding that GPT-3 shows such errors while ChatGPT and GPT-4 avoid them and perform hyperrationally, using the Cognitive Reflection Test and semantic illusions.
Large language models (LLMs) are currently at the forefront of intertwining AI systems with human communication and everyday life. Therefore, it is of great importance to evaluate their emerging abilities. In this study, we show that LLMs like GPT-3 exhibit behavior that strikingly resembles human-like intuition - and the cognitive errors that come with it. However, LLMs with higher cognitive capabilities, in particular ChatGPT and GPT-4, learned to avoid succumbing to these errors and perform in a hyperrational manner. For our experiments, we probe LLMs with the Cognitive Reflection Test (CRT) as well as semantic illusions that were originally designed to investigate intuitive decision-making in humans. Our study demonstrates that investigating LLMs with methods from psychology has the potential to reveal otherwise unknown emergent traits.