Wish Brockman from OpenAI posted elsewhere besides X. Until then I’ll just copy and paste these words of motivation:

One trick to making ML progress is repeatedly increasing the precision of the engineering and decreasing the number of mysteries being papered over – i.e. just focusing on improving your understanding of the system you’re working with.

I can’t tell you how many times I have gone over and over the AI in Subtxt, conveniently refining and shaving off the edges.

It takes some getting used to.