Article
Looki launches Loom, a life-understanding model: perception, experience, and long-term memory processed in three layers
Beating AI News: AI wearable device company Looki has released Loom, a multimodal model series that can organize everyday life recorded by cameras and microphones into long-term memories with time, content, and original evidence, for AI to retrieve at any time.
For example, a user has lunch with colleagues at noon, during which they discuss a project and finally decide to wait until Friday's data comes out before making a decision. Loom must not only recognize "eating" and "conversation," but also record the discussion content, the decision, and its reasons. When the project is discussed again a few days later, the AI can retrieve the record from that time.
Loom is divided into three parts:
Sense (perception): continuously determines what the user is doing. It uses a specially trained discriminative model to directly score questions such as "is the user eating" and "is the user talking," without generating a text description each time. The model also refers to the state from the previous period to avoid repeated fluctuations in activity recognition caused by turning around or occlusion.
Moment (experience): aligns images and sounds collected at different times, automatically determines where an experience begins and ends, and generates a description containing key facts, time, and corresponding evidence. It can segment experiences based on conversation content, rather than merely cutting video by fixed duration.
Memory: organizes experiences into daily, weekly, and monthly summaries and long-term memory. When answering questions, the AI can first check the summaries, then find specific experiences layer by layer, and when necessary review the original images or recordings to verify details.
At present, Loom still works through the coordination of multiple specialized models and a memory retrieval system, and plans to further integrate these capabilities into an end-to-end multimodal model in the future.
In official self-testing, Sense recognizes nearly 20 categories of daily activities, with an F1 score of 88.6 and an average judgment time of 0.51 seconds per judgment. Moment surpasses Gemini Flash in determining the start and end times of experiences, but the text descriptions it generates are still less accurate than the latter. Memory scored 79.0, 18 points higher than mem0, and the average number of Tokens processed per question is only 14.5% of the latter.
Loom will first be used in Looki's own wearable devices. It has not yet opened an independent model API, and the model weights have not been open-sourced.