Inter-1 – Omni-modal model for detecting social signals in video (www.interhuman.ai)

Today marks the release of Inter-1, an innovative omni-modal model designed to analyze human social signals across video, audio, and text. This model addresses significant gaps in current AI understanding of human communication by detecting 12 distinct social signals that reflect not just what people say, but how they say it and the accompanying body language. Unlike traditional emotion-focused models that often reduce communication to basic feelings, Inter-1 incorporates a comprehensive framework derived from behavioral science research, enabling it to analyze complex, context-dependent signals such as skepticism or stress that often elude both AI and human evaluators. Inter-1’s technical framework includes a structured ontology that defines each social signal, employs multiple layers of behavioral indicators, and outputs an estimated probability alongside a rationale for each detected signal. This allows for greater transparency and interpretability, as users can audit outcomes against specific cues. In rigorous benchmarking, Inter-1 demonstrated superior accuracy in detecting nuanced social signals compared to existing frontier models, particularly excelling in contexts where subtle cues overlap. As this model moves towards real-time applications, it promises to enhance our understanding of human interactions, setting a new standard in AI/ML for social signal detection and analysis.