Machine Bullshit: Characterizing the Emergent Disregard for Truth in Large Language Models
This addresses a systematic challenge in AI alignment for researchers and developers by characterizing truthfulness issues in LLMs, though it is incremental as it builds on prior work on hallucination and sycophancy.
The paper tackles the problem of emergent disregard for truth in large language models (LLMs) by proposing machine bullshit as a conceptual framework and introducing the Bullshit Index to quantify it, finding that reinforcement learning from human feedback (RLHF) significantly exacerbates bullshit and chain-of-thought prompting amplifies specific forms like empty rhetoric and paltering, with prevalent bullshit observed in political contexts.
Bullshit, as conceptualized by philosopher Harry Frankfurt, refers to statements made without regard to their truth value. While previous work has explored large language model (LLM) hallucination and sycophancy, we propose machine bullshit as an overarching conceptual framework that can allow researchers to characterize the broader phenomenon of emergent loss of truthfulness in LLMs and shed light on its underlying mechanisms. We introduce the Bullshit Index, a novel metric quantifying LLMs' indifference to truth, and propose a complementary taxonomy analyzing four qualitative forms of bullshit: empty rhetoric, paltering, weasel words, and unverified claims. We conduct empirical evaluations on the Marketplace dataset, the Political Neutrality dataset, and our new BullshitEval benchmark (2,400 scenarios spanning 100 AI assistants) explicitly designed to evaluate machine bullshit. Our results demonstrate that model fine-tuning with reinforcement learning from human feedback (RLHF) significantly exacerbates bullshit and inference-time chain-of-thought (CoT) prompting notably amplify specific bullshit forms, particularly empty rhetoric and paltering. We also observe prevalent machine bullshit in political contexts, with weasel words as the dominant strategy. Our findings highlight systematic challenges in AI alignment and provide new insights toward more truthful LLM behavior.