[6/21 7:00] UVU Deepfake public opinion manipulation research/AI robbery prediction 86.3%
AIニュース速報 · 2026-06-20 · 3 min
Show full episode description
6月21日 7時のAI速報をお届けします。 ▼ トピック 1. UVU ディープフェイク世論操作研究 https://kslnewsradio.com/utah/ai-deepfake-media-can-sway-public-opinion-as-effectively-as-real-media-uvu-study-finds/2308980/ 2. AI 強盗予測 86.3% https://techxplore.com/news/2026-06-ai-robberies-cities-accuracy.html 3. Nous Research Hermes Blank Slate Mode https://www.marktechpost.com/2026/06/20/nous-research-updates-hermes-agent-with-a-blank-slate-mode-that-pins-toolsets-via-platform_toolsets-cli-and-disabled_toolsets/ 4. Missouri S&T アリと鳥の群知能 https://news.mst.edu/2026/06/missouri-st-researcher-combines-lessons-from-ants-and-birds-to-improve-ai/
🌐 This transcript was automatically translated to English from the original.
June 21, 7:00 AM UVU's AI News Breaking News First: A Utah Valley University study found that AI-generated deepfakes can sway public opinion just as much as real media.In an online survey of more than 600 people, just 16% of people were able to tell the difference between real and deepfake videos, published Thursday by Utah Valley University's Gary Haber and Public Policy Research Institute. Furthermore, there was no statistical difference in opinion change between those who saw the real thing and those who saw the fake, making deepfakes as trustworthy or more trustworthy than the real thing.Researchers warn that anyone can use publicly available tools to create a convincing hacking video on an old student's laptop, and are concerned about the impact on elections of all sizes.Second news: An AI model has been developed that predicts robberies in multiple U.S. cities with 86.3% accuracy. Published in the Fukunishi City International Journal of Innovative Computing & Applications and validated with historical data from Los Angeles and Seattle.The GraphDaniComi network maps geographic relationships to identify risk points, and the Transformer module analyzes temporal trends, outperforming the strongest competing method by 83.2%.While it could help law enforcement optimize staffing, accuracy remains a challenge in areas with less crime data.Third News: NAUS Research has added a blank slate mode to the open source AI agent HARMIS that starts from a blank slate. At startup, it starts with a minimal configuration of only the provider and model file operations and terminal, and the user intentionally adds the necessary functions. Since the settings of the selected toolset are explicitly written to disk, even after an update command, unselected functions will not be enabled automatically. Designed for use in security-oriented operations, ensuring reproducibility in teams, and in training and auditing situations. We need a model that supports at least 64,000 tokens of context.Last news: Researchers at the Missouri University of Science and Technology have published a study that combines lessons learned from ants and birds to improve AI. One is ant colony optimization, inspired by how ants collectively find the most efficient route to food. The other is particle swarm optimization, inspired by how birds share information and collaborate. By integrating these two biomimicry techniques. We aim to realize AI with excellent adaptability and visibility, and relatively low power consumption. Swarm intelligence, born from natural swarms such as ants, bees, birds, and moths, is attracting attention as a new design guideline for AI. That's all for breaking AI news. Thank you for watching.