Too many people of my acquaintance place a lot of stock in Hanlon's Razor, specifically because their immediate thought after running it is "... But nobody could be that stupid."
Evil is not a great model for explaining bad results much of the time. Stupidity has same problem.
McKenzie's Razor and Shaving Cream: Don't attribute to evil what you can explain by emergent behavior in complex system.
You should aggressively update your estimate on whether a system is bugged or not by inspecting whether the system delivers sensible outputs given its inputs.
This is instrumentally useful for technologists because you probably spend a lot of your time around very smart people and very bugged systems, and therefore you will not prematurely rush to conclude "Hmm that there black box has smart people operating it; probably flawless."
• • •
Missing some Tweet in this thread? You can try to
force a refresh
There is something a little magical about giving your kids Dragonbox for the first time and watch them just intuit some of the rules of algebra, based on good game design.
Liam even raged at the app for not allowing him to use the communicative property of multiplication. He definitely doesn’t know those four words, but “If 1X is X and X1 is X then clearly it doesn’t matter if I drag the one onto the X or the other way!!!”
Liam: Why’s I get two stars?
Me: On the seventh step you waved your fingers a bit too much and caused a tap that the app thought was wrong. Which is fine because…
Liam: WE DO IT AGAIN.
Two weeks ago my buddy @fulligin pulled me aside and said "I have something to show you." That something was the Daylight Computer, which I was previously unaware of.
I've since bought one and am using it fairly extensively for reading. Vincent recorded some impressions:
The DC-1 has a whiff of magic to it, comparable to the first time using a Kindle.
(Other people profess to have this feeling from the iPhone/iPad the first time. I honestly don't remember those as being "my life now has a before and an after" moments. Kindle was that.)
The Kindle is a substitute for paper, via eInk. It's the best available way to buy books, a somewhat mediocre reading experience which acts like an inferior substitute for paper, and an absolutely gobsmackingly frustrating software/hardware artifact to use.
Matt Levine has a great piece today (well, you knew that) covering Goodhart’s law in the context of how ineffective it is to track mouse movements as a proxy for white collar productivity.
I think this is tempting thing for management to institute because there is a… suspicion.
That suspicion is not “so-called ‘email jobs’ are actually not productive at all.” This is believed by many people on the Internet. Those people are greatly miscalibrated. The world runs on email. Email causes physical results in the world. Coordination/communication are valuable
The thing which management believes is “I have a rather strong suspicion that, somewhere in my vast workforce of people doing white collar jobs, there are malingerers. Not merely people who are bad at their job. No. Some who are actively abusing me, and laughing about it.”
A memorable moment in new employee training for me was when the trainer said that most people working professionally in our field could not whiteboard out funds and data flow between all entities in a credit card transaction. She asked why I could, given no experience in industry
The answer I gave was that I had credit card processing accounts before and read the documentation and contracts carefully. This was true, but was a bit of a fib.
The real reason was I was curious about how PayPal worked in 2004 and had some time to kill.
You’d be surprised what you can learn by reading more of the Internet than anyone thinks is reasonable.
You’d also be surprised how many people consider themselves cogs in a machine, with a bit of knowledge of what their gears touch and then no real interest beyond that.
Once upon a time I wrote letters on behalf of people to banks. I don’t do that anymore, in general, but I still get asked sometimes. I have been told, but not tried much, that GPT is very good when you tell it to cosplay as me specifically doing this.
So today I got an email.
In “help the correspondent learn to fish” mode, I suggested they use ChatGPT to do this. But, since prompt engineering is a thing, thought I could take 30 seconds to write an example prompt.
This is not verbatim a letter I once ghostwrote on behalf of someone in exactly this situation, and you can see in the transcript that I had one or two thoughts. But otherwise, this is almost exactly what I’d write.
Took 30 seconds, and not the 15 minutes this typically did.
(If one doesn’t like the LLM part of it then sculptors exist, and decoupling their location from the building’s location is ~newly an option.)
(I suspect the guy I like in France who does really nice elves would be happy to use the usual workflow on the usual tools to deliver the usual STL file of a Greek god with the funny non-usual detail being the final print will not be 32mm tall.)