Multimodal Analysis - Live Data | AI in Academia - Dataspheres AI

Multimodal Analysis - Live Data MULTIMODAL THEMATIC ANALYSIS Every modality, one coded frame A 1h45m Senate hearing (diarized, 10 speakers) + six Stanford ...

Multimodal Analysis - Live Data MULTIMODAL THEMATIC ANALYSIS Every modality, one coded frame A 1h45m Senate hearing (diarized, 10 speakers) + six Stanford HAI & UNESCO panels (Gemini video analysis), coded into 8 themes with per-segment stances. Full write-up and live tables: dataspheres.ai/pages/ai-in-academia/multimodal-thematic-analysis-live Every chart in this deck reads its dataset live - the numbers update as coding continues. Corpus: Senate HELP hearing June 2026 (public domain), 5 Stanford HAI panels, 1 UNESCO talk. 31 coded segments and growing. Themes across the corpus Where the conversation actually lives Governance leads: evidence gaps, breaches, and FERPA-era law meeting consumer-grade AI. Bound to the Coded Segments dataset, grouped by theme, counting rows. Re-code the corpus and this chart moves. Stance by theme Pros and cons are not evenly distributed The optimism concentrates in pedagogy and labor; the caution concentrates in governance and equity. Both charts bound to the Theme Sentiment Summary dataset - left sums the pro column, right sums the con column. Overall stance The corpus is cautious, not hostile Balanced and con segments together outweigh pro - but almost every con comes with a policy ask attached. Pie bound to Coded Segments grouped by stance. The texture: witnesses argue pro on access and con on chatbots in the same testimony.