things

s12e57 · 27 September 2022

A Grab Bag

This issue meanders through nostalgia, computing infrastructure, disaster preparedness, and corporate accountability, loosely connected by the theme of systems and how they function—whether that's the beloved SCUMM engine powering a classic game or Tesla's proprietary supercomputer designed to train self-driving algorithms. The writer's real focus seems to be on visibility and transparency: contrasting the hidden regulatory compliance work at companies like Facebook with more explicit, public-facing processes like building permits, and using the building approval model as a deliberately clumsy provocation for how Big Tech might be forced to show its work more openly. There's a wry skepticism running through it all—appreciation for engineering ambition tempered by frustration that these systems operate behind closed doors, inaccessible to the public they affect.

457 words · 2 min read · 5 links · canonical link · JSON

0.0 Context Setting

It's Tuesday, 27 September, 2022 in Portland, Oregon, where right now it's 80f/26.6c.

1.0 Some Things That Caught My Attention

A grab-bag, this time:

  • Defining Moments: The SCUMM Bar is as short Lucasfilm blog post about that bit where Guybrush Threepwood enters the bar for the first time. Caught my attention because: there's that new Monkey Island game, but also, well, it's Monkey Island.
  • Tesla provided details about its Dojo computing architecture at Hot Chips 34[1], caught my attention because: wow, lots of reasons. They're building (already have) a supercomputer that would be within the top 5 of the top500, they're doing private exascale computing, they hired a bunch of people from IBM, ARM and so on and threw out a bunch of junk that x86/ARM processors use and the whole thing still runs PyTorch. I mean, I don't love Tesla, and then Space Karen goes and throws money at this kind of thing, where "this kind of thing" is "design a new computing architecture so that you can train a computer to drive a car", and that's before you even get into "can you even do that well". But I guess we get a new computing architecture (locked away, in private IP, not for consumer use), as a treat?
  • Heard about the book Disasterology, by Dr. Samantha Montano about emergency disaster management, which... well, it's interesting. Stuck it on my library pull list, at least.
  • I was at a Trusted Design workshop this morning and had cause to remember something I mocked up in 2018 that I would uncharitably call a "provocation", which is to say: what if Facebook had to behave externally, publicly, in the same way that building developers behave and issue notices and run consultations? I am reasonably sure I've written about this before. If I have, I think the difference this time is remembering that there's a whole bunch of regulatory compliance work that goes on inside that's hidden from public view, so counter to the (naive) notion that Facebook people just sit around and do whatever they want, fuck around and find out, move fast, break things yadda yadda. No, the difference this time is in noticing the making-explicit and ugly. I agree that directly mimicking the building planning permission process doesn't make sense because you could do it better but that's also part of the (sigh) provocation. Of course you could do it better, this is simply illustrating one bad way of doing it.

Okay, that's it for today! Still getting back into the swing of things.

How are you doing?

Best,

Dan

  1. Inside Tesla's Innovative and Homegrown "Dojo" AI Supercomputer, Timothy Prickett Morgan, The Next Platform, 23 August, 2022 ↩︎