米拉·穆拉蒂的 Tinker 提供实时 AI，但它能击败 OpenAI 吗？
米拉·穆拉蒂的思维机器实验室预览了“交互模型”，该模型以 200 毫秒间隔处理音频和视频，以实现实时处理人类打断。这一方法挑战了 OpenAI 和 Anthropic 主导的轮流聊天界面，预示着 AI 模仿人类对话动态的未来。
mira murati emerged from a year-and-a-half of silence to preview 'interaction models,' a new class of ai interface developed by her startup, thinking machines lab. unlike the turn-based chat paradigm established by chatgpt, these models process continuous audio, text, and video in 200ms intervals to capture the texture of human conversation, including interruptions and pauses. murati described this as a first step toward a more natural human-computer interaction, declining to set a specific release date for the technology. beyond the product reveal, murati reflected on the 2023 openai governance crisis, arguing that the industry relies too heavily on individual virtue rather than structural checks and governance to prevent catastrophic errors.