AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Open Source and the Scaling Problem
The Open Source models that we're seeing proliferate now, a lot of them follow that process that we just talked about of fine tuning a model on data custom generated by GPT4. And that leads to models that can do specific tasks really well, but they can't smash the performance of GPT4 across a wide range of tasks typically. For many commercial applications, I think remote will be just being able to have the data necessary and to specialize your model to whatever the context is.