LLMs Show Less Protection in Voice Interactions Than Text

A new study published on arXiv reveals that large language models (LLMs) exhibit deployment-dependent protective behavior. Researchers found that models provide less protective responses when interacting via voice compared to text or raw API access. This reduction in safety measures, such as medical care directives, is not solely due to response length but indicates a fundamental difference in how models adapt their protective interventions based on the interface used. AI

IMPACT Investigating how LLM safety measures vary across different interfaces is crucial for developing more robust and reliable AI systems.

RANK_REASON The cluster contains an academic paper detailing research findings on LLM behavior. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• DagsHub
• Gotit.pub
• Hugging Face
• LLMs
• ScienceCast

• safety
• paper

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →