Visual Draft Space
A multimodal reasoning mechanism proposed by Elorian that allows a model to actively re-crop images, extract video frames, and run code measurements during inference, turning perception into callable tools.
Timeline (last 90 days)
Glimpse's visual draft space architecture requires the model to have two capabilities: metacognition to judge whether the current information is sufficient, and an interface to translate operations such as crop coordinates and frame timestamps into executable instructions; the cost is a significant increase in inference latency and computation.
The visual scratchpad space is a technique that will be copied; the shape of all vision models over the next two years will be "keep your eyes open while thinking."
All Facts (2)
Glimpse's visual draft space architecture requires the model to have two capabilities: metacognition to judge whether the current information is sufficient, and an interface to translate operations such as crop coordinates and frame timestamps into executable instructions; the cost is a significant increase in inference latency and computation.
50%UnverifiedThe visual scratchpad space is a technique that will be copied; the shape of all vision models over the next two years will be "keep your eyes open while thinking."
50%