Every Major LLM is a 1-Box Smoking Thirder
What are modern LLMs' decision theoretic tendencies? In December of 2024, the @Caspar Oesterheld et al. paper showed that models' likelihood of one-boxing on Newcomb's problem is correlated with their overall decision theoretic capabilities (as measured by their own evaluation questions) as well as with their general capability (as measured...
Mar 2410