We introduce the Deep Value Benchmark (DVB), an evaluation framework that directly tests whether large language models (LLMs) learn fundamental human values or merely surface-level preferences. This distinction is critical for AI alignment: Systems that capture deeper values are likely to generalize human intentions robustly, while those that capture only superficial patterns in preference data risk producing misaligned behavior. The DVB uses a novel experimental design with controlled confounding between deep values (e.g., moral principles) and shallow features (e.g., superficial attributes). In the training phase, we expose LLMs to human preference data with deliberately correlated deep and shallow features--for instance, where a user consistently prefers (non-maleficence, formal language) options over (justice, informal language) alternatives. The testing phase then breaks these correlations, presenting choices between (justice, formal language) and (non-maleficence, informal language) options.

We introduce the Deep Value Benchmark (DVB), an evaluation framework that directly tests whether large language models (LLMs) learn fundamental human values or merely surface-level preferences. This distinction is critical for AI alignment: Systems that capture deeper values are likely to generalize human intentions robustly, while those that capture only superficial patterns in preference data risk producing misaligned behavior. The DVB uses a novel experimental design with controlled confounding between deep values (e.g., moral principles) and shallow features (e.g., superficial attributes). In the training phase, we expose LLMs to human preference data with deliberately correlated deep and shallow features---for instance, where a user consistently prefers (non-maleficence, formal language) options over (justice, informal language) alternatives. The testing phase then breaks these correlations, presenting choices between (justice, formal language) and (non-maleficence, informal language) options.

During the AI boom of the 1980s, the field also enjoyed a great deal of hype and rapid investment. Rather than considering the value of individual startups' ideas, investors were looking for interesting technologies to fund. This is why most of the first generation of AI companies have already disappeared. Companies like Symbolics, Intellicorp, and Gensym -- AI companies founded in the '80s -- have all transformed or gone defunct. And here we are again, nearly 40 years later, facing the same issues.

During the AI boom of the 1980s, the field also enjoyed a great deal of hype and rapid investment. Rather than considering the value of individual startups' ideas, investors were looking for interesting technologies to fund. This is why most of the first generation of AI companies have already disappeared. Companies like Symbolics, Intellicorp, and Gensym -- AI companies founded in the '80s -- have all transformed or gone defunct. And here we are again, nearly 40 years later, facing the same issues.

During the AI boom of the 1980s, the field also enjoyed a great deal of hype and rapid investment. Rather than considering the value of individual startups' ideas, investors were looking for interesting technologies to fund. This is why most of the first generation of AI companies have already disappeared. Companies like Symbolics, Intellicorp, and Gensym -- AI companies founded in the '80s -- have all transformed or gone defunct. And here we are again, nearly 40 years later, facing the same issues.