David, a Bostonian in Tokyo.

https://mastodon.mit.edu/@djl

Retired translator hiding out under a rock in Tokyo, Japan. Interests includes photography, jazz guitar, juggling, Materials Science, Go, Japanese literature, Comp.Sci., and AI.
Programming experience in Fortran, Lisp, BCPL, C, C++, Python, and assembler on IBM, CDC, DEC, and Intel architectures.

Public Key

npub1s35zmlgr9aka5jxtd9x98lclcfw7wjy9nqedfdksszdnt3h2gr0seameq8

NIP-05 Address

[email protected]

Profile Code

nprofile1qqsgg6pdl5pj7mw6fr9kjnznlu0uyh08fzzesvk5kmggpxe4cm4yphcpzemhxue69uhhyetvv9ujuurjd9kkzmpwdejhgqg5waehxw309aex2mrp0yhxgctdw4eju6t04kmr7q

Publishing to

relay.primal.net

relay.damus.io

nos.lol

Author Public Key

npub1s35zmlgr9aka5jxtd9x98lclcfw7wjy9nqedfdksszdnt3h2gr0seameq8

Show more details

Last Notes

2026-08-29 02:39:17 CEST

- reply

Ha! I've been there. The first 12 shots from my 2022 folder are from a walk to and back from that temple. It's a bit of a ways from the station. The Tokyo shot is from the top of an open-to-the-public comms facility of some sort, and the second pagoda shot is from the same thingy. (Click on a thumbnail for the photo itself.)
https://pbase.com/davidjl/photos_2022

2026-08-26 16:35:59 CEST

- reply

"AI architecture is not an interest of mine..."
That's sensible. But. What LLMs do is quite simple.
LLMs are exactly and only _statistical text generators_. Albeit fancy ones.
The AI bros claim that statistical operations on enormous text databases can generate causal reasoning and thus intelligence.
That breaks the first law of statistics: correlation is not causation.
Thus LLMs can't possibly do what's claimed.
#AI #generativeAI #statistics

2026-08-17 12:47:34 CEST

- reply

Some other points are that (a) the AI companies are paying multiple otherwise underemployed good mathematicians gobs of money to spend full time mucking with their LLMs, (b) a good mathematician can tell the LLM which steps are flaky (and can use Lean to check some steps quickly).
And (c) we aren't told how many failures there were (or how much said failures cost to do) and (d) we aren't told how much the successes cost to create.

2026-08-11 13:36:56 CEST

- reply

I suspect it's worse than you think. The zeroth level problem is language that falsely/incorrectly anthropomorphizes their programs. When you read "(the AI) chose to try to ..." you should smell BS-dense hype.

2026-07-08 03:08:26 CEST

- reply

Yes. That was rather weird, being about the only clear thing you could hear.
1972 was an odd year. Prior to 1972 there were anti-war (and other) protests all the time. But in 1972 it all stopped. And it wasn't just the US, it was worldwide (well, apparently France and Japan as well, dunno about other places).
My theory was that the US switched the draft to a lottery system in which you largely knew whether or not you were going to be forced to spray Asian children with napalm or not.

2026-07-08 02:17:35 CEST

- reply

You can piano a tuna if you just put it on the ground next to MIT's Baker House dorm. (From the roof of which the kiddies drop a piano every once in a while.)
https://www.youtube.com/watch?v=FoDsA1WnMAU

2026-06-24 02:42:50 CEST

Ed Zitron is at it (i.e. being brilliant and spot on nailing it) again. Sheesh. Boring, guy. Get a life.
"Why would everybody agree to do something so stupid? Why would everybody act so crazily?
It’s simple: the tech industry has completely run out of ideas, and all that’s left is a cargo cult that hasn’t had a human experience since 2015."
https://www.wheresyoured.at/cargo-culture/

2026-06-16 14:28:36 CEST

- reply

"The math is simpler for a hollow hoop than for a solid ball, so I'd start there."
I smell a frictionless ping-pong ball in a vacuum. Or is it a perfectly frictionful ping-pong ball in a vacuum?
The latter, I guess...

2026-06-01 16:49:52 CEST

- reply

Because of the order in which your wife asked. Ask again in the other order and it'll probably give a different answer.
Friends don't let friends use LLMs.

2026-06-01 04:16:12 CEST

- reply

I couldn't find my copy of Perceptrons so I grabbed the latest copy with the late 80s, early 1990s additions. Their main point was that it's necessary to prove that your model can actually do X if you think/claim that it can do X. It seems to be mostly down-in-the-weeds details of those proofs.
I'm currently reading Jerry Fodor, whom I think largely has the right ideas wrt. AI, theory of mind, and the like. But that's not much help for a deep learning boook...

2026-06-01 03:51:14 CEST

- reply

I'm not aware of who did the lambda stuff in C++, I just read about it and played with it a bit after it was standardized and implemented in Mickey$oft C++.
After my two years at AT&T I threw a few stink bombs at the C++ standardization folks. There was some noise from Japan arguing for some sort of subset standard, both for microcontrollers and for users who didn't want to learn the whole thing. I liked that idea but it got flamed royally. (This was early 1990s.)

2026-06-01 03:38:18 CEST

- reply

Hilariously, the C++ community understood what Steel and Sussman had done, and added it into into C++. Correctly and thoroughly. Good work, guys!. But. With the ugliest, most obscure, most unreadable syntax possible. Did I mention that C++ is insane? Yes, I did. And it is.
But there's only one standard.

2026-06-01 02:45:30 CEST

- reply

"In the beginning, there was Caveman Lisp. We didn't understand function calling, so where your free variables got looked up was random. Steele and Sussman fixed this problem, but no one understood what they had done."

2026-05-20 21:48:02 CEST

Understatement of the decade:
"Although you may have pain for a few days after getting Shingrix, it will be less severe than the pain from shingles and its complications."
- CDC (US)
I assure you, there are very few things in this world anywhere near as severe and debilitating as shingles pain.
My doc checked my antibody levels (like 10 years ago) and said they were more than adequate. Ten years later, they weren't.
Get the damn shot.

2026-05-15 03:38:15 CEST

- reply

He must have been foreseeing ADA...

2026-04-25 03:00:21 CEST

- reply

The other day the Tenseijingo column in the Asahi started with a quote from Natsume Soseki's novel, "I am a Cat", which has the cat complaining about the summer heat.
The author looked up the summer temperatures for the year the novel was written. The high temperatures were...
30 degress Centigrade.
Last summer, 10,000 people were hospitalized in Japan for heat-related problems. The first responder work of getting such patients to a hospital is brutal. F'ing saints, they is.

2026-04-03 18:31:34 CEST

- reply

Welcome to retirement. Where keeping track of the day of the week is the hardest thing we do.
That's why Fenster Freitag and Monochrome Monday are so important.
Lovely photo; great space and angles.

2026-03-12 21:24:13 CET

- reply

Huh? My friend with an iPhone is already running models locally on said iPhone.
The somewhat larger point is that there's no economic model that makes the current AI datacenter frenzy anything other than financial suicide. For somewhere in the free to $20 a month range, people will use them. But running a bevy of halucinating agents that might delete your data (after stealing it) for serious money isn't something sensible folks will be doing.

2026-02-24 03:24:08 CET

- reply

There's a new bio of Hokusai out. By Isozaki Yasuhiko it's titled "The Life and Artistic Drive of Katsushika Hokusai. (5000 Yen at Amazon.jp)
"葛飾北斎の生涯と芸術意欲" by 磯崎康彦
There's a review in the January 10 Asahi Shinbun which I've clipped for my own nefarious purposes. Yell if you'd like a scan of it before I chuck it.

2026-02-19 01:47:58 CET

- reply

Hey! Keep posting these. Thanks.
If you are in town this month, the Ohta Memorial Museum is having an exhibition of Ukioe featuring middle aged men (Ojisan, not Ojiisan; go figure). The Asahi newspaper reports that it's crowded even on weekdays, and most of the visitors are women. Go figure.
https://www.ukiyoe-ota-muse.jp/ukiyo-e-ojisan-festival-eng/

2025-12-12 02:04:14 CET

- reply

Checking Wiki, it seems that Joan would have been in Boston 1959 and 1960 or so. I would have been around 8 then, so before my time, but just when dad was starting up the violin repair thing. Dad's violin repair business customer set was dense of MIT physics types, but I was only aware of them much later than that.
Oops: we didn't overlap much at Yale.

2025-12-11 12:14:57 CET

- reply

My dad*, whose last day job was safety officer for Harvard's chem department, was in the instrument repairing scene in the Boston area, and was visiting someone who did guitar work, and met your cousin Joan there, back in the day when I was first listening to Mimi.
She struck him as being rather intense.
Thanks for the thermo pdf book, by the way!
*: https://pbase.com/davidjl/image/121610792

2025-12-11 01:11:03 CET

- reply

Drat! I was thinking that one of his cousins singing masters of war would be a kewl response here, but neither appears to ever have sung it. Sigh.
Oh, well. Here's Mimi and friend singing Bold Marauder. (Richard and Mimi Farina remain on regular rotation here: some of the best music of the 60s.)
https://www.youtube.com/watch?v=QWELgzRLBl8

2025-09-05 03:36:06 CEST

- reply

I read frantically in grades 4-6, pretty much all the scifi the Boston Public Library had.
I read frantically in high school: the "hippy canon" of the time (1984, catch 22, Hesse, the Russians, the French, Tolkien, etc.)
I didn't read much in college. (MIT will do that to you.)
Learned Japanese, and now mostly read fiction (both light and literary) and lit. crit. in Japanese. Math I read in English. Stinkbombs on the internets I throw in English.

2025-07-06 14:53:03 CEST

- reply

Lovely use of 28mm, like the last few.
(I'm switching from 24 to 28, and haven't figured out 28 yet, so these are doubly appreciated.)