Uncertain Updates: December 2024
I’ve been busy!
First up, the book’s website is up! You can check it out now at fundamentaluncertainty.com.
Second up, the second draft is coming along nicely. I completed a major revision of Chapter 1 and am nearly done rewriting Chapter 2.
I didn’t really like the original version of Chapter 2. I was still figuring out the book’s purpose and style, and it contained the wrong material. The new version does a better job of making its points without so much flailing around.
I’m still flailing around trying to get it finished, though! I’m broadly happy with its content and structure, but it still has some rough patches I’m trying to smooth out. I might have gotten it done by now, but writing website copy was a surprisingly time consuming activity, so I’m only just now returning to it.
Claude-assisted everything
Speaking of writing, Claude has rapidly become my stalwart companion in writing and many other activities.
I’ve been using Claude for a while, but at the end of October I left my job (I expect to start a new one in January). With more free time on my hands, I’ve used it to really figure out how to use Claude effectively. Hardly a day goes by that I don’t talk to him.
Claude’s helped me do all kinds of things. He helps me write code. He helps me find answers to questions faster and better than Google can. And he even helped me make a cool playlist for an event—he knows a surprising amount about modern music!
He also helps me write. It’s a bit hard to explain the mindset shift required to work with an AI a partner, but that is really how I think of Claude now: a partner. My current workflow for the book looks like this:
I create a project to group together our book-related chats.
I link a Google doc with the current draft of the book to the project.
I start a chat for each major task, like revise Chapter 2 or write website copy.
I go back and forth with Claude, showing him things I wrote, asking him to rewrite things, asking for ideas about what I should include, etc. Basically anything I’d ask if I was writing together with another person, I ask Claude, but with the added benefit that I never try Claude’s patients and he’ll work whenever I’m ready to.
Some things Claude is pretty good at:
fact checking claims
rewriting bullet points into prose that approximately has my voice
copy editing
taking free writing and turning it into more polished prose
I have to watch out for slop! Claude will slip into slop mode if I don’t keep him tightly focused on the task at hand.
He also only kind of understands my voice, but will try to revert it to the mean. I usually have to edit everything he writes to make it really sound like me.
I also push back a lot. Claude kind of updates based on this, at least within a chat or project. We’re slowly figuring out how to work together better.
My takeaway is that I’m like 5x more productive than I am without Claude. The speed and quality of what I write goes up tremendously with Claude’s help. As one person put it to me recently, if you’re not working with an LLM partner, you’re already behind the productivity curve.
NB: I curiously didn’t get Claude’s help to write this issue of Uncertain Updates! I think it’s because I see this as a place to talk to you directly, rather than as polished writing.
Twitter
I keep having fun on Twitter. The last two weeks have been onsite interviews, so I tweeted a lot less, but I still have some threads to link!
https://x.com/gworley3/status/1858910742432673840
https://x.com/gworley3/status/1860860230579720691
On a related note, I might be spending too much time on Twitter, because I’ve started shitposting on Less Wrong:
Holidays
As has been tradition for 11 years now, this Friday I will celebrate the solstice with my rationalist friends in Berkeley. I’m even speaking in the program this year! Come join us if you can.
Whatever holidays you celebrate or don’t around this time of year, I hope you have a great end to 2024 and an even better start to 2025. See you in the future!