The scientific method becomes genuinely rewarding.

So what does a language model do?

Can we make language models follow behavioral rules without altering weights?

What is it like to be ChatGPT?