I'd like to see considerable expansion on this topic, in areas beyond generating domain knowledge training data. I don't know if you've done much work with small embedded models, but they will be the workhorses of my primary effort; I'll need data targeted at training them to focus tightly on very specific tasks, of which each model will only be required to learn two or three at most. I'm not sure how your insights would help with that.
Also, teeny-tiny text on scaled-down screen shots is quite difficult for some of us to read (impossible on mobile); perhaps including examples in normal text font instead would be more inclusive...
I'd like to see considerable expansion on this topic, in areas beyond generating domain knowledge training data. I don't know if you've done much work with small embedded models, but they will be the workhorses of my primary effort; I'll need data targeted at training them to focus tightly on very specific tasks, of which each model will only be required to learn two or three at most. I'm not sure how your insights would help with that.
Also, teeny-tiny text on scaled-down screen shots is quite difficult for some of us to read (impossible on mobile); perhaps including examples in normal text font instead would be more inclusive...