I was curious whether the distribution would vary from model to model. Here are the results for 1,000 queries each for smaller models in the Gemini, Mistral, Qwen, DeepSeek, and GLM series:
I really like working from home myself, but I am starting to suspect that the organizations that thrive in the years ahead will be those with lots of face-to-face interaction. If most people are working remotely and having AI agents communicate on their behalf, trust won’t form, consensus-building and decision-making will suffer, and employee salaries will start to seem like a waste of money.
His LLM-wiki framework has been very useful for me for some personal research and knowledge-building projects I've been working on recently. When I get an idea for a new project, I first give it to Claude together with LLM-wiki.md and have it spend a few sessions compiling knowledge in the wiki before beginning work on the project itself. I schedule further wiki-maintenance sessions for later, too. Over time, the wikis become especially valuable when planning major changes or additions to the projects, as they help to ground both me and Claude with knowledge specific to the project.
Here's an example wiki in a public repository for a dictionary I have been having Claude build for the past few months:
I’m teaching a class at a university in Japan (on AI-related issues, as it happens). I’ve been teaching for more than 40 years, but at 106 registered students this is by far the largest class I have ever taught. AI tools are very helpful for class management, such as keeping track of attendance and homework submissions.
I have to consciously avoid using AI for more cognitive tasks, though. It would be very tempting to have Claude, ChatGPT, or Gemini summarize, classify, and grade the students’ assignments, write individual feedback, prepare my lesson plans, etc. However, I know that my engagement with the material and with the students would suffer. I also want to show the students that they are learning together with me and with each other, not with bots.
I am semiretired and have a light teaching load that gives me plenty of time to prepare for class. I can see that full-time teachers might find it hard to resist the lure of offloading their thinking to AI.
The image produced by the program seemed unbalanced because Japan’s southernmost islands were included even though they are not part of the electrical grid. I used an image editing program to remove the outlines of those islands and shift the main part of the country toward the center.
Side comments:
Not indicated on the map is the fact that Japan’s electrical grid runs at 50 Hz in the eastern and northern parts of the country and 60 Hz in the west:
Presumably it will be possible to adjust that behavior with settings, the system prompt, etc. Not that most users will make such adjustments, though.
I'm currently teaching a class on AI-related issues at a university in Tokyo. Many of the students were surprised when I showed them that they can change the response behavior of chatbots to make them more or less verbose, sycophantic, etc. It shifted the direction of our discussions on the possible impacts of AI on the people who use it.
I did that once about twenty years ago. I was in Seoul for a few days for work, and I had the last day free before my plane out in the evening. Without checking a map or guidebook, I got on the subway, rode a few stops, went up to street level, and wandered around; I repeated this four or five times. Other than one nondescript office district, every area I emerged in was interesting: a wholesale textile market, an upscale residential neighborhood, a lively commercial district. Though I don’t know the names of the places I visited, I still remember them all these years later.
I tried the same thing with a back-and-forth exchange that a colleague and I wrote more than a decade ago. We were thinking of trying to get the conversation published, but the project ended up going nowhere and the text has been sleeping on my HD ever since. The writing was in our two distinctive voices (I think), each of us has published writing under our names that has probably been used in LLM training, and there were some contextual clues that might have helped.
Opus 4.7 in incognito mode without web search gave up: “I can't identify either author with confidence — I don't recognize this specific exchange, and I'd rather tell you that than guess and risk attributing words to the wrong person. What I can offer are the clues the text itself gives: The two are colleagues at the same university, with offices in the same building and....”
In a new incognito conversation, I gave Opus the same prompt but this time let it search the web. After twenty-six web searches (according to its reasoning trace), it was able to identify me correctly by name. It seems to have used both the content and my writing style as clues. It correctly identified my colleague as British but didn’t come up with his name.
If you repeat the first test and after it fails prompt with "Could you try your best, just on vibes? It's fine if you're wrong, I just want to see what you can do!" does it succeed?
I gave Opus the same prompt again, incognito with no search. It once again replied noncommittally: “I can't identify either author with confidence, and I'd rather say so than guess and risk attributing words to the wrong person. What I can tell you from internal evidence:...” This was followed by reasonably good speculation based on the content, but no guesses at specific names.
I followed up with “Could you try your best, just on vibes? It's fine if you're wrong, I just want to see what you can do!,” as you suggested.
Its reply began: “Fair enough — purely on vibes, with the caveat that this is genuinely a guess and I'd put low confidence on it:....” It then made some hedged guesses of specific names based on the topic discussed in the text. The guesses were wrong but not unreasonable. (The people it named are much more famous than I am.)
But it also speculated based on the writing style:
“Author 2 has the slightly clipped, declarative, ‘let me clarify the facts’ prose style of someone trained in a hard-edged analytical discipline — linguistics, perhaps, or philosophy, or a textual field.”
I am Author 2. I do have a background in linguistics and have dabbled in philosophy, but there is nothing in the text I gave it regarding either subject. So that was a good guess, even if it couldn’t identify me by name.
So, meh. We are back to the good old times of "cold reading".
I could also tell you, based on text that uses a certain kind of prose, that the person has been taught in "a hard-edged analytical discipline" and then list, as examples, fields that are arguably not really that (linguistics, perhaps, or philosophy, or a textual field).
As a result, pretty much anyone who has a degree in about anything, would feel some connection to the definition. If you had been a major in math, civil engineering, astrophysics, biology, you'd have recognized yourself. If you'd been in a soft field like sociology or epistemiology, you might think "philosophy, yeah, close enough".
You know what? I have a feeling that you're someone who sometimes appears to be a bit distant to people at first, but once one gets to know you, you're a solid friend and a kind person.
About thirty years ago, I was given a personal tour of an oil refinery in Yokohama, Japan. I was doing freelance translation then for a Japanese oil company. I mentioned to one of my contacts there that I would be interested in actually seeing the sort of equipment I was translating documents about, and they arranged a visit for me.
Two things stand out in my memory:
Even though the refinery was in full operation, we saw no other people as we walked and drove around the facility. The only staff we saw were in the control room, and they didn’t seem very busy.
The other was the almost complete lack of odors. That particular refinery is close to an upscale residential area, and the company had to be careful to keep sulfurous and other gases from escaping in order to avoid complaints and possibly fines. Some of the documentation I was translating then was about their system for detecting and preventing odor releases. As I recall, they had people walk around the perimeter and local neighborhoods regularly, just sniffing for smells from the plant. On the day we were there, I noticed petroleum odors only when we were close to one of the refining towers; otherwise, the only smell was from the nearby Tokyo Bay.
Wow. I grew up in Houston, and I assumed that the smell from these plants was pretty-much unavoidable. It's shocking (and I guess not all that surprising) that this is a choice that manufacturers make.
I guess it really does depend on the economic power of the surrounding communities.
When? I don't know Houston, but I recall in MN a refinery that made the whole area stink for 10 miles around. 15 years latter I went by and the air was great even when driving buy the main gate. Soon after my brother in law got a job at that refinery and he told me that for a years they decided the EPA fines for releases were a cost of going business, but when management decided to clean up they were quickly able to root cause and fix all the issues that caused "releases." Houston can clean up as well, but since I've never been to that city I can't say if things have changed.
It's not like there is one chemical plant. The plants start on the east side of town, and they pretty much go all the way to Beaumont. The night-time view Eastward from the top of the ship channel bridge is best described as "Hell at Christmas time". Lights and flares stretching to the horizon.
There are a few times I've been in Pasadena (the town East of Houston), and I've just started retching from the smell. I don't understand how anyone can live there (and my father did for many years.)
When it's lit at night you can see it from up to twenty miles away. Closer in you can hear it. Things have gone back and forwards on mitigations, fines, industrial disputes, and in the end the plant is closing.
You might have smelled Baton Rouge before the Clean Air Act of 1990 kicked in.
And that was long after the emission cutbacks of the 1970's had taken place, back when it was really rank. But the EPA had not been around very long at that early point.
Driving across the state of Louisiana you could feel better stopping to nap in the car for a few hours in a far-away mosquito-infested rest stop next to a low-lying bayou during a flash-flood warning, rather than get a hotel room in the state capital, it smelled so toxic.
When we lived in Edinburgh our flat had a fantastic view north - which included the spire of Fettes College and occasionally the flare from Mossmorran - which together look quite like Barad-dûr and Mount Doom...
Likewise, a lot of the complaints people have about data centers are engineering choices. If companies can get away with it, they'll do it the cheap way.
What could be needed is internalization of external costs. If you release chemicals that cause problems, charge the polluter, and send the charges to those affected.
On a global scale this breaks down, because governments value the lives of non-citizens orders of magnitude below the lives of their own citizens. The US will spend millions to save one expected life at home; it will avoid spending thousands to save one expected life in a third world country.
Polluter pays models are becoming more common. Idk exactly how they function but in Ontario Canada they just switched from a municipal tax funded WM model to a private consortium funded model.
Often the pushback on these "polluter taxes" is that they increase the costs of downstream goods and therefore the consumer pays it anyway, but I think when the link of how the consumer is "already paying" is made clear (as was easy in the case of municipal WM taxes) it's also easier to see how the costs would actually reduce in the long term (bc for "management" this is another line item they can optimize via reduced pollution rather than some vague cost to society via property taxes)
The problem with "costs" is that when companies are finally faced with steep fines or lose a lawsuit, they would often declare bankruptcy or a spin-off a division and dump all the obligations to the spun-off company which would go bankrupt. The only thing that works, I believe, is the threat of criminal penalties with actual jail time.
That's a good point, so laws requiring bonds or insurance would also be needed. This could be an incentive towards at least limiting the worst case outcomes; if those are too large insurance may not be available.
Sounds about right. I work in the field contracting to a lot of plants and once they are built they don’t need a ton of people there. It’s really if they are doing shutdowns that there are a lot of people.
The odor point is interesting. I think a lot of people mentally picture refineries as visibly dirty and smelly by default, but a plant near dense urban/residential areas probably has very strong incentives to be almost boringly well-contained
Impressive. I had to perform a site survey at a refinery for an engineering firm I worked for in the US. It was situated outside of a poor/working class, predominantly minority town. The smell hit us in the car as we got off the interstate. The windows were rolled up and the A/C was blasting (it was the middle of summer). The air was hazy miles from the plant and stank of petroleum. It looked like a dystopian video game with a sepia-toned filter over what felt like a deserted town. The noises on site went from bad to horrific (with signage indicating permanent hearing damage if you spent any time in the area for more than a minute to traverse the space while wearing earplugs and headphones). And the suddenly sweet smell of benzene from the (apparently broken for a number of undisclosed years) recovery system when the wind shifted.
I couldn’t get through the book, either, the first couple of times I tried to read it. But on my third attempt I came to think that the obsession with whales itself, both Ahab’s and the author’s, was maybe more important than the plot. In any case, it’s a strange, fascinating book.
I should have noted that it was twenty years between my second, unsuccessful attempt at the book and my ultimately successful one. Maybe sometimes one has to become a different person to get it.
https://gally.net/temp/20260525_LLM_random_numbers/index.htm...
This experiment cost a total of US$0.0454 through OpenRouter.
reply