Booth 21-25 | AI Data Management Zone | Tokyo Big Sight
Most improvements to AI systems happen before deployment: better training data, better fine-tuning, better RLHF. Once the model is out in the world, you generally get the performance you trained
The dominant model for AI-assisted research has been batch processing: submit a query, wait for the system to work through it, receive a result. The interaction is asynchronous, the feedback loop is slow, and the human is largely passive until the output arrives. A paper from early 2026 presents a
