A new research paper evaluates four leading real-time voice AI systems—OpenAI's GPT Realtime 2, Google's Gemini 3.1 Flash Live, and Alibaba's Qwen3.5 Omni Plus and Omni Flash—finding they often fail to act on vocal cues like emotion or sarcasm, despite being able to perceive them. This "emotional intelligence gap" means the systems prioritize the literal words spoken over the tone and delivery, leading to inappropriate responses in critical scenarios such as customer service or financial transactions. While prompting can offer partial improvements, the research suggests caution when using these systems in contexts where vocal delivery is crucial. AI
IMPACT Highlights a critical gap in current voice AI, suggesting caution in applications where vocal tone and emotion are vital for understanding.
RANK_REASON Research paper published on arXiv detailing findings about AI model capabilities.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
