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
Here’s a problem that doesn’t always make it into model deployment conversations: AI models degrade over time, even when you don’t touch them. Not because the model changes. Because the world does. This is concept drift — when the relationship between the inputs a model was trained on and the
