Reflections on InkHaven
It’s (finally) over!
If you’ve been wondering why I’m suddenly blogging every day… well, it’s about to stop! I decided last minute to join InkHaven, i.e. commit to blogging every day for the month of April.
This was a somewhat questionable decision because I have a lot of other shit going on. For instance, I’m starting a non-profit. But I figured:
I could keep it to one or two hours a day max; I mostly succeeded at this, although I did get sucked in sometimes, and I think probably spent almost four hours on the longest one.
I consider myself a good writer, but I have never been able to write fast, except when commenting on things, I figured this would be good practice. It was! I really got over this block to a huge extent! I only editted my posts sometimes, and very minimally, and I think time-boxing both the writing and editting is probably good practice.
I’ve been meaning to write more about my views on AI, and in particular, why I think it’s an urgent crisis that may require (and warrant!) drastic actions like getting rid of AI computer chips. I made some progress here, but not as much as I would’ve liked. It’s hard to sacrifice quality for things you really care about!
So for my final blog, I figured I’d do a quick retrospective. What have I learned? How do I feel about the whole thing? Here’s a list of observations:
I feel positively, but kind of lukewarm. Why is this, I wonder? I guess I was hoping for more. My greatest hope (which I wisely decided not to make a proper ambition) was that I would get most of my core thinking on AI risk down in writing. I think I got more like… 10-20%?
I feel like I have a lot to say, and I’m surprised by how little of it I feel like I said and also how hard it was many times to figure out what to blog about that day.
A lot of posts ended up feeling a lot bigger than 500 words. I spent more time on the project than I intended.
As always, writing stuff demands clarity and precision that highlights one’s confusions. This was useful.
I was often pulled to write about things other than AI, but largely avoided the temptation. I have very mixed feelings about this. I think it would have been a much more rewarding experience in a lot of ways to write about whatever I fancied. I might have actually done more of that if I had another place to put it -- this blog is about AI, after all… But I think I was also held back by the concern that it might be better to be more private and keep my public persona more “professional”. I really don’t know if that’s the right approach, and I don’t like it, so maybe I shouldn’t be doing it.
At times, I lost track a bit of what I’d said in previous blog posts. I’ll probably have to go back through and read them all at some point. I hope there’s some good content in there that I can repurpose for something more cohesive and comprehensive.
I wish I’d tracked my time a bit better… how close was I to spending 1-2 hours (max)? I know I did it in <1hr a few times for sure…
Oh hey, look, I’ve made it to 500 words!
But seriously, if you’ve stuck it out, I appreciate it! I expect I will keep blogging some, and that the quality will go up now that I can take my time more with the posts.
One thing I keep turning over: you say the barriers are political, not technical. But the political barrier is that AI is people's prestige and livelihood—researchers, companies, whole national strategies are built on it. Where I'm studying, this is especially stark: the research culture rewards competing for position far more than asking whether the work should be done at all, and qualification often matters less than the race. So asking the field to set this down voluntarily seems to run into the very incentives that make the risk dangerous in the first place.
That's why I'm genuinely unsure which path is faster. Is organizing the public to reject AI more tractable than it looks, because those incentives only bind the insiders and not the public? Or does that same incentive problem make technical work—on regulation, verification, control—the more realistic near-term lever, even if you think "steering and hoping" is ultimately a losing bet?
And it's exactly that competition-over-qualification culture I'm trying to step outside of—which is why what I want from you is your insight, not just a place in the race. I'd really like to hear how you weigh the two paths.
Glad you're going to keep blogging—you mentioned wanting to get more of your thinking on AI risk down in writing, and for what it's worth, the posts you've already written are part of what reached me.
I'm a 4th-year CS undergraduate in South Korea. The real danger of AI is being hidden behind its capabilities and its visible benefits—most people around me haven't grasped it. These systems can just as easily be turned toward authoritarian control, corporate tyranny, or individual abuse, and the concentration of power you write about is the part that frightens me most. AI research and grad admissions are overheated here and elsewhere, and I was about to walk that same road without thinking hard about it. But when I try to forecast where this goes, I keep landing in the same place: I don't trust human motives to hold, and I think it won't be long before AI is heavily weaponized. I want to help stop that, and I want to contribute.
I would genuinely like to study and do research under you. I know you're on leave from your faculty job and not taking students right now—so rather than asking for a position, I want to ask directly: is there any path for someone at my stage to earn that kind of opportunity down the line? I'd take time off before graduating to study or pursue a PhD with you if that ever became possible. Short of that, I'd be grateful to keep exchanging insight with you over time—and to know whether there's any way to be useful to Evitable's mission now.
I want to research AI in order to make it safe, and where necessary, to help stop it.
—Yohan, Sungkyunkwan University