MIT

MIT study: AI models in home surveillance inconsistently detect crime, recommend police calls, and show demographic biases in decision-making

MIT's Entrepreneurship JetPack uses AI to rapidly answer startup queries, streamlining development. It's trained on a 24-step framework and provides instant research.

MIT's Crystalyze AI reveals crystal structures from powder samples. It predicts lattice size, composition, and atom arrangement with 67% accuracy on new data.

Co-LLM allows general AI to collaborate with expert models, improving factual accuracy. It uses a "switch variable" to defer to specialists at word level, enhancing efficiency.

MIT's ScribblePrompt AI speeds up medical image analysis by 28%. Trained on 50,000+ simulations, it highlights structures via user scribbles across scan types.

MIT study finds 70% of AI training datasets lack licensing info, 50% have errors. New tool aims to improve transparency and dataset selection.