What It Takes to Ship a Small Model

Accuracy is subjective. It's defined differently by every customer, even for the same use case. For large models, that's manageable. You can prompt and iterate. For small models, you have to get it right before you train. In this interview, Liquid's CTO Mathias Lechner talks with COO Jeffrey Li about precisely defining the use case upfront, customization beyond fine-tuning, and ongoing calibration as data shifts in the real world. Jeffrey argues the two things hardest to automate are the taste required to define what a model needs to do and the taste required to train it well. Subscribe to follow every interview: https://www.youtube.com/@liquid-ai-inc Careers at Liquid AI: https://www.liquid.ai/careers