Yash Patel on the History and Future of AI Model Training

Turning a restaurant menu into a food-delivery storefront requires more than reading its words. In this case study, the model also had to follow rules about attaching modifiers to items and distinguishing add-ons from ingredients.

The speaker describes having people correct the model's menus, then comparing generated outputs with those corrected examples during training. That comparison supplied a measurable error signal to optimize. The mechanism makes the company's definition of a correctly structured menu part of training, rather than relying only on general image understanding. The account describes the workflow without quantifying its improvement on held-out examples.

“These models today are not really like that.”

Yash Patel

“RL is kind of this like eval maxing machine.”

Yash Patel

“General models sort of set the floor, but in order to set the ceiling, you need to go and build, train models, create these specialized systems in order to differentiate yourself from all your competitors.”

Yash Patel

“My honest take is like scaling Transformers is working and as you know, there's a very simple recipe to be able to make these things smarter and better.”

Yash Patel