Andrew Dai argues that multimodal models can recognize visual patternsA vision-language model jointly processes images and language so it can describe, answer questions about or act on visual information. while failing to reason reliably about counts, spatial relationships and changes over time. Andrew Dai uses chess, board-game and object-trackingComputer vision enables software to analyze visual inputs such as images or video and extract information about their contents or structure. examples to distinguish recognition from deliberate visual reasoning.
Andrew Dai describes Elorian's approach to multimodal data, synthetic trainingSynthetic data is information generated to represent examples or situations rather than collected directly from the events it describes., architecture and visual chains of thought. RoboticsEmbodied AI perceives and acts through a physical or simulated body while interacting with an environment., construction monitoring and CAD illustrate intended applications rather than independently verified production capability, and no safety guarantees are established.
Watch on YouTube




