The "prompting" ability you're referring to is called "instruction following", and here are some descriptions of it. https://openai.com/blog/instruction-follow

The "prompting" ability you're referring to is called "instruction following", and here are some descriptions of it. https://openai.com/blog/instruction-following/ https://openai.com/blog/instruction-following/ I think the differences are more in the training data used, than in the nature of the model itself. So you could probably train your own instruction-following model on top of this raw 20B model.