AI advancements focus on brain-interface alignment and private on-device processing

Researchers at the Korea Advanced Institute of Science and Technology (KAIST), in collaboration with Microsoft Research Asia, have developed a brain-computer interface technique called Neural Value Alignment (NVA). Using electroencephalography (EEG), the system can detect brainwave responses when a person perceives an AI is pursuing the wrong goal or taking unexpected actions. This technology aims to address goal-action ambiguity by allowing AI to recognize unspoken user dissatisfaction and correct its behavior in real time.

In a separate development, Conway Research has secured funding from investors including Andreessen Horowitz, Khosla Ventures, and Hummingbird VC to develop Underdog. Underdog is a private AI assistant designed to run entirely on personal hardware, such as iPhones and Macs, without sending data to external servers. The company utilizes custom inference engines optimized for Apple Silicon, claiming performance speeds up to 4.5 times faster than Apple’s MLX framework. The product lineup includes the small-scale Woof model and the larger Underdog 27B, which is intended to compete with frontier models while maintaining local privacy and agentic capabilities.