MiniMax researcher Olive Song and Together AI engineer Dan Fu discuss the release of an open-weight modelAn open-weight model makes its trained parameter values available so others can run or adapt it under the release's licence. and the work needed to serve it. Olive Song describes multimodal training and applications involving images, video and computer use.
Dan Fu explains how inference teams adapt kernels, attention, caching and quantizationModel quantization represents AI model values with fewer bits to reduce storage, memory use and often inference cost. to an architecture, then improve performance after launch. The shift from short chat turns to long agent sessions changes which parts of the serving stack matter mostAI compute infrastructure is the hardware, facilities, networks, power, cooling, storage, and software used to train and run AI models..
Olive Song describes reinforcement-learning environments for longer tasksReinforcement learning trains an AI system to choose actions using feedback about the results of its earlier choices. and the need to distinguish genuine improvement from benchmark gaming. Both speakers discuss how open models and infrastructure can develop together.
Watch on YouTube
