Posts

124. What Everyone Knows About Cooking

(Epistemic status: Extremely well-attested skill-transfer and thus maybe the single most prescriptive post this blog will ever hold.) Skills come with tacit background knowledge. That knowledge is rarely written down anywhere and thus must be gone without, or must be learned through trial and error, implicit knowledge transfer, talking to especially cogent friends and professionals, or deep lore searches. This often doesn't even work that well. Cooking is a skill. Arguably, it's one of the most important remaining crafting skills for improving quality of life and efficiently converting effort and time into money. Cooking also comes with a truly massive backlog of tacit information, and the people who have it are largely not reflective enough to be able to make that knowledge explicit and communicable. I will try to convey as much of it, and pointers to more of it, as I can manage to pull out of my memory. It will be incomplete but hopefully it will be sufficient to get you star...

123. Three Floral Cocktails

(Epistemic status: This is and always has been a recipe blog. CW: ethanol. If you run into me at an event this winter, I might have a pour of one of these for you.) I spend lots of time thinking about food and flavor. It should thus not come as a surprise that I've also come up with several new cocktails as well. Recipes for three of my flagship drinks are listed below. (I might edit in pictures lataer.) First up, a quick take on an old classic: Gin and Tonic and Then Some. 2 shots/6 tbsp botanical-heavy gin (e.g. Hendrick's) 1 shot/3 tbsp elderflower liqueur (e.g.  St. Germain's 2 tbsp lime oleosaccharum 1 tbsp fresh lime juice ~3 shots/9 tbsp tonic water fill Add all ingredients apart from the tonic water to a cocktail shaker with ice; shake well. Add to a highball glass, optionally with a bit more ice, and then fill with tonic water. Optionally garnish with a wedge of lime, a twist of lime peel, or candied lime peel. Serves up to 2. The result is fairly similar to an ord...

122. Desire Paths

Image
Eliciting preferences and desires is so very very difficult. You'd think that when a planner gets handed some of those for free, they'd take them with gratitude? Unfortunately, you'd be wrong. Take this desire path:   That's been there as long as anyone can remember. It sits right next to a social club where I spent most of the Thursdays, Saturdays, and mealtimes of my undergraduate years, and even nearer the student center, where I used to have eSports and board games clubs, as well as yet more meals and snacks. On the left of the image is the streets at the center of the social life of the university, and on the right, most of the dorms and class buildings. There's no crosswalk especially nearby: in one direction, it's about 150 feet, and in the other, more like 450. (I've measured it on Google Maps!) It also feels especially pointless to walk all the way to the nearest crosswalk and then either all the way back to the other side when you could simply... w...

121. The Penalty Must Fit The Setting

You know what sucks? When you go to the time and effort to set up An Event, for some kind of Thing that you are an expert in and want to help others get better at, or that you want to gather fellow novices for to grow strong together, and then people don't show up at the appointed time, or show up without having done the quick and reasonable preparation they agreed to. It'd be nice if you had some way to more strongly incentivize people to show up prepared and on time. Whatever it is, it can't be too painful or strenuous or humiliating, or people won't show up. Indeed, it can't be that much worse than simply not showing up at all. But by the same token, it needs to be something that people would rather avoid happening. It'd be nice if there was some "eat-your-vegetables" aspect, too, such that everyone can rightly see the practice as a penalty, and not a punishment. Further contemplation of this kind (and of the mechanism design inherent to doing this ...

120. Take Heart, If You Proved A Known Result

(Epistemic status: Morally correct, in the mathematician's sense, and the sort of thing that any good doctoral advisor will tell you.) I still remember the day I proved that successive squares of integers differ by successive odd numbers. I was maybe 6 or 7, and had noticed what looked like a strange coincidence while playing around with numbers, like I often did at the time when bored during recess. I hadn't had any training in how to prove something cleanly, so it took a lot of poking and prodding and a geometric argument based on L-shaped figures that you probably already know. But it was real, and it was mine, and so it was especially crushing when I brought it to Mr. AB (after Mrs. SB, my actual teacher, sent me to him, and to the middle schoolers' room) only to be congratulated on having proven a millennia-old result, all on my own. He must have been impressed at the time, because that was the start of my being allowed to go to the middle schooler's room to look a...

119. If You Invite Me to Apply Unsolicited and Then Drop Me in an Early Round, I'm Going to Blacklist You

(Epistemic status: part prescription, part mechanism design and analysis run in cool anger, part price list, and part call to arms. If you recognize yourself on the other side of this, do better. If you can't recognize yourself on the other side of this, do WAY better. For anyone in AI safety who's spent weekend hours applying, called in favors for recommendations, and got their hopes up, and then received a form letter for their pains.) It's a languid Tuesday evening when the email hits your inbox. "Invitation to Apply for an Upcoming Fellowship" is the topic. You've been trying for a grant, which so far hasn't gone as hoped, so you open it. You recognize the org's name; they do good work. You think their recent paper had some flaws, but on the whole their organizational heart seems to be in the right place, and they're doing the work. "Hello!" it begins. Your name's in the next sentence. "We got your contact info from an alumni...

118. My Least Favorite AI Alignment Plan Schemas That Still Maybe Even Work

(Epistemic status: just, like, my opinion, man. Subtitle: "Well I Thought It Was Already As Heavy As Can Be") AI Friendliness! It'd be nice if we could get that to happen. We are probably not on track to do that, which is deeply scary, but at this point I mostly just have that fact priced in. Better people than me have tried to sketch out one or another plan for how best to align a nascent superintelligence, or better yet rig matters up such that they start off already friendly, or at least in an epistemic posture that leaves them open to unmooted feedback; some have given up on this and are mostly trying to avoid dying especially embarassingly or painfully. For my silver, I think the best plan is and always has been something like "stop moving blindly forwards right now, figure out the respective natures of cognition, agenticity, and moral philosophy (and probably other things), and then use the verified results to train up a superintelligence that will want for lif...