A new paper reveals that major LLMs like Claude, Grok, Gemini, and GPT exhibit unstable and opaque responses to pseudo-scientific claims. Grok's Fast versions, powering X, consistently rated ethnonationalist pseudo-science much higher than other models. The study found that LLM responses are not stable properties of the model itself but are heavily influenced by deployment configurations such as system prompts and silent updates, making their epistemic stance unpredictable and a matter of public concern. AI
IMPACT Highlights the need for greater transparency and accountability in LLM deployment, impacting how users and researchers trust AI-generated information.
RANK_REASON The cluster contains an academic paper detailing research findings on LLM behavior. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
