Mira Murati的Thinking Machines Lab發布「交互模型」以促進人機協作

mira murati’s thinking machines lab has unveiled a new class of multimodal ai models, termed 'interaction models,' which utilize camera and microphone inputs to natively understand messy human communication. this technology allows the ai to adapt in real-time to interruptions, tonal shifts, and clarifications, departing from the standard architecture of transcribing speech for text-based processing. murati explicitly contrasts this collaborative philosophy with the superintelligence pursuit of firms like openai and google, asserting that keeping humans in the loop creates better futures. the lab also offers 'tinker,' an api product launched in october 2025 that allows researchers to fine-tune open-source models with custom data. despite raising billions, the company has not yet publicly released the interaction models, though demo videos showcase their intent-aware capabilities.