This is a thoughtful new review of the “interactive explanation” milieu: distill.pub/2020/communica…. I’m a friend of the format—I’ve written articles like this myself—but I worry it’s trapped in a limited framing, selling short the potential of computational representations.
Here's my crux: The Cartesian plane was not invented to disseminate mathematics, or to make math more engaging. It was invented to help *do math*.
The same point can be made about John Snow’s cholera map, Feynman QED diagrams, and our other most powerful representations.
If you create an interactive representation which amplifies original research, then you can often *also* use it for dissemination, journalism, etc. But if your design goal is “communicating to others,” it’s very unlikely that the representation will amplify original research.
This is what we see in practice, with almost all these articles: representations designed to introduce an audience to an idea, and no more. Many other “explorable” authors have confided uncertainty to me about the impact of this work. With love: I suspect they’re right to worry.
A more aspirational goal is to help decision-makers make complex decisions by “shipping the model”. This aims closer to the mark. But it’s usually lip service: pieces like The Atlas of Redistricting (projects.fivethirtyeight.com/redistricting-…) are more for idle play than serious analysis.
Even Jupyter-based “executable books” erect huge barriers between embedded figures and ongoing use. Try extracting a model from an article-notebook to use in “real” work. It’s rarely easy, and I think that illustrates the medium’s priorities.
In summary, there’s a reason these articles usually feel like one-offs, and that the field doesn’t seem to be accreting: the representations are rarely deep enough to stand on their own and build upon each other. They’re too often for showing, not for thinking. Bret put it well:
(This topic really deserve a full essay—there are many other important problems and opportunities in this space—but my queue is full enough that tweets are all I can afford for now…)
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Some notes on @thesundaylight, a sort of… artificial skylight?
It feels good to spend the day in daylight. But that doesn’t happen when I’m working inside. Especially in the winter. So, for years, I’ve used super-bright LEDs while I work. They quite noticeably improve my energy and mood.
The main problem is that existing ultra-bright lights aren't great for home use. Options include: 1. industrial lamps (ugly, poor CRI) 2. 20+ high-CRI bulbs 3. film lighting (huge, often have fans)
All these will make your space feel like a warehouse unless you add bulky soft boxes .
Sunday's approach: it shines a very bright light *up*, at a large circular diffuser, which casts the light into the room.
The diffuser has a coating which really does make it feel like the sky: it's vivid blue, and the spotlight's manipulated reflection looks like the sun.
The Diamond Age's "Primer" has long been edtech's most canonical shared vision. I feel the field is haunted by it. It's wrong in many important ways—but we haven't articulated a better one, so we cling to it. I want to transcend it.
The vision has been so sticky because there's a *lot* that's right about it. In particular, the emphasis on immersion, responsiveness, a sense of assurance, and the emotional experience. andymatuschak.org/primer/
But the concept is foundationally flawed.
Principally, its authoritarianism: the Primer has an agenda of its own; Nell isn't told what it is; the Primer is not interested in Nell's goals or values. It's supposed to develop subversives, but Nell spends her life trapped in a box.
U Chicago offers a four-year discussion based course on “The Great Books”, available online to anyone. I just finished year one, and I’d definitely recommend it! graham.uchicago.edu/programs-cours…
Some notes:
It’s really remarkable how much well facilitated discussion improves my understanding. Well-facilitated conversation would often show me quite forcibly how shallowly I’d grasped the text. Others’ understandings were almost always different in interesting and enriching ways.
At least in my sections, the other students were mostly older or retired professionals (lawyers, teachers, managers, etc). My ordinary life isn’t really full of conversation with these folks, and I really enjoyed exposure to perspectives I wouldn’t be getting otherwise.
I'm not sure if I'm part of the intended audience for Anthropic's safety statement, but unfortunately it left me more uneasy than before I'd read it.
Conspicuous in its absence: discussion of incentives. Particularly in light of the rumored $5b raise. (con't)
Incentives are a big problem here! The founders, funders, ICs, alumni… the org-as-entity itself. In every case, these entities will face enormous net incentives to accelerate. If the claim is that they’ll *not* do this, despite that, then that claim needs extended justification.
What exactly are the terms necessary to raise $5b in venture funding, and then to make technical alignment the top priority?
Anthropic's career page notes: "We offer competitive compensation with significant amounts of equity." How should employees contend with equity vs. acc?
Given the recent surge of interest in AI for education and evocations of The Young Lady's Illustrated Primer, it's fun to ask: what are we to make of the Han girls raised by the Primer becoming the Mouse Army, rather than individualistic subversives like Nell?
One explanation is given by Dr. X: "We lacked the resources to raise them individually, and so we raised them with books. But the only proper way to raise a child is within a family. The Master could have told us as much, had we listened to his words."
In this view, the Han girls turned out differently because they had an AI raising them instead of a ractor like Miranda; and they lacked supportive influences like Harv, the Constable, and Miss Matheson’s Academy.
But another read is that the Han girls were actively sabotaged.
My sight reading practice has been totally transformed by Sheet Music Direct’s subscription service. It’s all-you-can-eat sheet music, a nice addition to IMSLP for modern stuff. Most songs have “easy” arrangements ~around my sight reading level. Tons of new music every night!
Also nice to use it in conjunction with RCM’s popular selections list, since it’s helpfully broken into fine-grained grades. Can usually read laterally within any of the volumes mentioned, in addition to the specific pieces. rcmusic-kentico-cdn.s3.amazonaws.com/rcm/media/main…
I’ve noticed that if I sight-read an “easy” arrangement of a piece, I can often sight-read some of the “real” arrangement afterwards—a bit of scaffolding… I guess it gets the basic harmonic structure in my head.