I’ve got a few projects in the oven, but nothing’s baked quite yet. So this newsletter it’s links-only, organized in descending order of Kevin-commentary.
Two tips for maximizing your information intake/enjoyment:
In Gmail, create a filter for the word “unsubscribe”, label matches with “newsletter” (or whatever) and have them skip the inbox. Now you never need to worry about cluttering your inbox and you can sign up for a newsletter every time you read a thoughtful article. (No newsletter? No problem, use Blogtrottr to convert any blog RSS feed into emails.)
Install the video speed controller browser extension and give it hotkeys (I use the square bracket keys) so you can easily adjust playback speed for on any video playing on any website. This especially helpful for slow-talking YouTubers (30m woodworking tutorials become 10m) and speed-running “getting started with our enterprise software solution”-type videos.
Now that you have my secrets, you too can curate wonderful, link-filled newsletters. (Obviously, let me know so I can sign up!)
Okay, onto the links:
DeepMind has solved protein folding. They have a PR blog post, but I recommend going straight to Yannic Kilcher’s explainer video for an overview of the problem space, DeepMind’s 2018 algorithm, and speculation on their (as-of-yet unpublished) latest algorithm.
See also this structural biologist / machine learning researcher’s 2018 sociological commentary about how a handful of nerds from an advertising company ate the lunches of the much larger academic and pharmaceutical research communities. (Again, this was in 2018 when DeepMind merely doubled the usual rate of improvement — this same researcher described the last week’s results as “I think it’s over”.)
Aside: I’ve been interested in the folding problem since my undergrad physics thesis, where I developed a fast protein structure search/similarity algorithm: Encode structures into a fixed alphabet via unsupervised clustering of Voronoi diagrams, then run spell-check check.
While I never published in a journal, I did benefit by learning early just how much even single-node MongoDB databases liked to corrupt themselves, which helped me avoid the tire-fires which burned so many of my peers through the mid-2010’s.
Solar power is getting very cheap. This was on my qualitative radar, but only recently did I learn the numbers: Utility-scale solar costs are now 20% what they were in 2010. In many places, it’s price-competitive with the operating costs of coal power plants.
This isn’t interesting because your electric bill gets cheaper; it’s interesting because energy is an upstream cost behind pretty much all goods. (See OtherLab’s Super Sankey diagram for an overview of US energy flows.)
For example, electricity is a third the cost of aluminum, so solar power getting cheaper could make it feasible to build more things from a lighter, stronger material than before. (I know some of y'all run hedge funds and otherwise allocate substantial capital — check out this proposal for building solar next to Australia’s bauxite mines and please remember to invite me to your yacht parties after you’ve dominated the global aluminum industry.)
Even more interesting are the opportunities related to dynamic costs. Due to production variability (i.e., clouds), solar capacity is usually over-provisioned, which means that on sunny days electricity might be literally free. (Or even negative, when it’s cheaper for a fossil fuel plant to pay you to take their extra electricity for an hour than it is for them to shutdown and startup again.)
This is the framing that’s worth keeping in the back of one’s mind: Sure, some process might normally be too electrically-intensive to do all the time, but is it worth doing when the electricity is free? E.g., creating fertilizer directly at the farm. I’m sure the efficiency looks terrible compared to Haber–Bosch, but if you’ve already got panels on the barn and you can’t sell the electricity back to the grid, why not use it to pull nitrogen out of the air?
“[O]ne person, working alone in their bedroom, was able to build a capability which would allow them to seriously compromise iPhone users they’d come into close contact with.” Just skimming through this technical opus is a reminder of the unfathomable complexity of our familiar computing devices.
That this finding — the flagship phone of the world’s largest company can be wirelessly taken over — barely makes Hacker News, much less mainstream outlets, suggests to me that we all knew this sad state of affairs in our hearts already.
This is bizarre: We live in a culture where I’ve gotten multiple physical letters and phone calls imploring me to have replaced (at manufacturer’s full expense) my car’s passenger windshield-wiper motor (which in some rare circumstances may overheat and fail), but, yeah, the device which mediates all my social relationships (with humans) and binding legal/financial relationships (with businesses and states) will not just give up all my secrets but can be actively subverted against me by any rando within wireless range, yep, computers are just like that sometimes, oh well.
The Archimedes drive is a planetary gearbox, except using smooth rollers instead of gears. No gear teeth means no backlash, which is an extremely useful property for building precision mechanical systems. Normal roller drives are difficult to manufacture because the tolerances must be exact (too small and the rollers won’t make contact; too big and they destroy each other / cannot be assembled). The obvious-in-hindsight innovation of the Archimedes drive is to hollow out the rollers so they become squishy enough to work with. (The spring constant to be tuned for specific materials and applications by the amount of hollowing). See Nick Parker’s writeup for more detailed enthusiasm.
Thanks to reader Mike McDonald for saving me from a mess of exploding oil-filled watches (mentioned last newsletter) by pointing me to this forum post discussing the need to compensate for the ~100x thermal expansion coefficient difference between metal and oil. Based on the dates and initials, it looks like that forum poster actually did get around to patenting their “thermal expansion compensator for liquid-filled watches” (spoiler: closed-cell foam).
“We have developed a custom 3D printed nozzle for our lasercutter. […] there are a lot of resources for designing DIY laminar flow nozzles for water, and that’s for one reason: super soakers.” (The nozzle minimizes burn marks.) These folks are ballers, see also their capacity-constrained Voronoi diagrams, and the resulting jewelry and sweet custom kicks.
A podcast interview with a satellite engineer. As required by law for all electronics projects, satellites too have a blinking LED so its designers can tell if it’s working (via telescope). Also, with a large enough radio dish it’s possible to detect whether your satellite’s crystal oscillator is running!?
I was personally attacked when a so-called “friend” suggested that my wireless keyboards might be a forever project.
China won’t just cancel your IPO, they’ll throw shade at you on the state-run bedtime podcast. Then again: “[Jack Ma] knew what he was doing and got what he wanted.”
“The framing of ‘cognitive bias’ privileges the global knowledge game at the expense of a richly situated rationality”
“If we study how they fixed goto, will it teach us how to make more usable concurrency APIs?”
“A micro-supercapacitor was fabricated on a coconut by the multiple lase method”
“Other socialists from elsewhere around the country were, as socialists in America typically are, left-wing intellectuals detached from practical governance. And they made fun of these Milwaukee guys for constantly talking about their public works rather than overthrowing capitalism. But what the Milwaukee guys got was that if you want non-market provision of stuff, there comes a time when you actually need to go do the stuff.”
“A Chain Just Cut Through A Capsized Cargo Ship Filled With Cars And The Process Is Fascinating”
A cube chock full ‘o reaction-wheels that can balance, jump, and otherwise move in an adorable/terrifying fashion.