-
Visualise and/or animate the effects of git commands - a warning that this blew up when I tried using it to visualise a simple
git rebase -
Sapling is Meta/Facebook's source control system, a fork and rewrite of Mercurial which they've started to open source. The fact that it interoperates with existing git repositories makes me optimistic that one day, our profession's long nightmare will be over.
-
On controlling the language we use to talk about machine learning - I don't hold out much hope for this, as AI is already firmly entrenched as a modern myth, but it's still worth having conversations about.
-
I first noticed this long blog post from someone in the AI risk community about GPT back in September - it predates ChatGPT - and didn't read it at the time because I guess I've been reinforced to assume that doing that would not maximise my utility function, but I decided to update my priors and give it a go. (The history of nerd culture is the history of ways in which we think of more and more sophisticated ways to avoid saying the three words "I was wrong".) It's a slog and will be incomprehensible unless you've already learned a lot of the LessWrong dialect, but it's interesting to see a community with a heavy investment in their fantasy-league AI grapple with what large language models are capable of. Some uncharitable thoughts: spinning out very long slabs of prose which fit into a conceptual framework but which are unmoored from anything like lived reality is exactly what, from my outsider's point of view, the AI risk community have been doing themselves for years, amd large language models means that they no longer even have to write the blog posts.