pbw

This will happen for real. But then a JIT-like process will generate regular code for efficiency, on the fly in the background.

Rather than “the book explains how bread is made” say “the sheets of paper which make up the book have ink in the shape of letterforms which correlate with information about how bread is made”.

I’m going to give this URL to Claude, ask it to propose uses of state charts in my codebase. Yes, it already has this in its training data, but I find giving it a URL brings it to top of mind.

This is only 33 years after I took a networking class and learned all about IPv6 and the IPv4 address space crisis.

There's certainly a risk that an individual will rely too much on AI, to the detriment of their ability to understand things. However, I think there are obvious counter-measures. For example, requiring that the student can explain every single intermediate step and every single figure in detail.

A two-hour thesis defense isn't enough to uncover this, but a 40-hour deep probing examination by an AI might be. And the thesis committee gets a "highlight reel" of all the places the student fell short.

The general pattern is: "Suppose we change nothing but add extensive use of AI, look how everything falls apart." When in reality, science and education are complex adaptive systems that will change as much as needed to absorb the impact of AI.

This sounds good, but I wonder if AI has changed the calculus on conflict resolution. It can not only chase down the conflicting changes, but also read those commit messages and PRs to divine intent. It might be that git is "good enough," given we have AI.

I have mixed feelings about the "Do X in N lines of code" genre. I applaud people taking the time to boil something down to its very essence, and implement just that, but I feel like the tone is always, "and the full thing is lame because it's so big," which seems off to me.

SO was built to disrupt the marriage of Google and Experts Exchange. EE was using dark patterns to sucker unsuspecting users into paying for access to a crappy Q&A service. SO wildly succeeded, but almost 20 years later the world is very different.

Gary Marcus, ha! He's generally not entirely wrong, but boy, is he annoying.

This is food for thought, but horses were a commodity; people are very much not interchangeable with each other. The BLS tracks ~1,000 different occupations. Each will fall to AI at a slightly different rate, and within each, there will be variations as well. But this doesn't mean it won't still subjectively happen "fast".

There's an HDR war brewing on TikTok and other social apps. A fraction of posts that use HDR are just massively brighter than the rest; the whole video shines like a flashlight. The apps are eventually going to have to detect HDR abuse.

Whether this exact approach catches on or not, it's turning the corner from "teaching AIs to develop using tools that were designed for humans" to "inventing new tools and techniques that are designed specifically for AI use". This makes sense because AIs are not human; they have different strengths and limitations.

Workday is a disaster, at least the version we have.

To capture the individual transistors on a modern CPU, you'd need an image tens of terabytes in size, and it'd have to be captured by an electron microscope, not an optical image. And even that wouldn't let you see all the layers. Some of the very old CPUs, I'm not sure what resolution would be required.

"I created the Alphanum Algorithm to solve this problem. The Alphanum Algorithm sorts strings containing a mix of letters and numbers. Given strings of mixed characters and numbers, it sorts the numbers in value order, while sorting the non-numbers in ASCII order. The end result is a natural sorting order."

https://web.archive.org/web/20210207124255/http://www.daveko...

And you think the comparison to book reviews is equally bad? Both are from GPT-5.

Essentially, Jenson's complaint is "When I ask an LLM to 'summarize' it interprets that differently from how I think of the word 'summarize' and I shouldn't have to give it more than a one-word prompt because it should infer what I'm asking for."

Yes, GPT-5's response above was not shortening because there was nothing in the OP about Plato's Cave. I agree that Plato's cave analogy was confusing here. Here's a better one from GPT-5, which is deeply ironic:

A New Yorker book review often does the opposite of mere shortening. The reviewer:

* Places the book in a broader cultural, historical, or intellectual context.

* Brings in other works—sometimes reviewing two or three books together.

* Builds a thesis that connects them, so the review becomes a commentary on a whole idea-space, not just the book’s pages.

This is exactly the kind of externalized, integrative thinking Jenson says LLMs lack. The New Yorker style uses the book as a jumping-off point for an argument; an LLM “shortening” is more like reading only the blurbs and rephrasing them. In Jenson’s framing, a human summary—like a rich, multi-book New Yorker review—operates on multiple layers: it compresses, but also expands meaning by bringing in outside information and weaving a narrative. The LLM’s output is more like a stripped-down plot synopsis—it can sound polished, but it isn’t about anything beyond what’s already in the text.

LLM's can shorten and maybe tend to if you just say "summarize this" but you can trivially ask them to do more. I asked for a summary of Jenson's post and then offer a reflection, GPT-5 said, "It's similar to the Plato’s Cave analogy: humans see shadows (the input text) and infer deeper reality (context, intent), while LLMs either just recite shadows (shorten) or imagine creatures behind them that aren’t there (hallucinate). The “hallucination” behavior is like adding “ghosts”—false constructs that feel real but aren’t grounded.

That ain't shortening because none of that was in his post.

We need to get the male’s genetic material into the female’s body. How many redundant copies should we send. 100? 1000? A voice in back of the room: 50 million.

This is cool but I feel like typing speed and vim skills are going to play less of a role in overall development speed as AI use increases. But certainly it won’t hurt to type fast, even if it’s mostly typing prompts.

Another datapoint is working earlier eras sound bad to me: punchcards, assembly, COBOL, FORTRAN. Yes I suspect those people had a blast.

I also started in the 1990’s and agree the evolution has been as you describe it. It does highly depend on where you work, but the tightly managed JIRA-driven development seems awfully popular.

But I fall short of declaring the 1990s or 2000s or 2010s were the glory days and now things suck. I think part of it is nostalgia bias. I can think of a job I spent 4 years and list all the good parts of the experience. But I suspect I’m forgetting over a lot of mediocre or negative stuff.

At any rate I still like the work today. There are still generally hard challenges that you can overcome, people that depend on you, new technologies to learn about. Generically good stuff.

By all means “organize together to place demands on your employers”. I didn’t say don’t do that. But there are 24 hours in a day — maybe strive to be good at your job AND organize instead of doing just one or the other?

“I'm 100% sure nothing will ever improve.” Nothing? Ever? Brainwashed?

Companies will always try to capture the productivity gains from a new tool or technique, and then quickly establish it as the new standard for everyone. This is frustrating and feels Sisyphean: it seems like you simply cannot get ahead.

The game is to learn new tools quickly and learn to use them better than most of your peers, then stay quietly a bit ahead. But know you have to keep doing this forever. Or to work for yourself or in an environment where you get the gains, not the employer. But "work for yourself" probably means direct competition with others who are just as expert as you with AI, so that's no panacea.

Is there a similar feature available for SQLite? Will there need to be a totally new extension for every DB, or is there a shared portion?

I'm the OP, and I think of it as more "completely and utterly inevitable" than "needed." Given our personalities and history, the promise of curing all diseases, along with many other promises, will compel us forward. But whether or not we'll look back and say it was the right move, I don't think anyone knows for sure.

Their experiments started in 2021 and the change was made in 2025? This makes me think AI writing code for us will only speed things up so much.

The board can and should be friendly with the p-corp owner. Almost all the time they are going to be green light everything. They are really just a "sanity check" (literally).

Now you could say how do we "make sure" the board acts when the time comes? Stands up to the owner? Maybe we don't. We put the mechanism in place, and if the board fails to stop the owner, then it didn't work in that specific case. And the world will know that. But as long as it works "most" of the time maybe that's enough.

Also I forgot, apart from a board a big thing might be reporting. Your p-corp activities would have much more stringent reporting requirements compared to a private individual. You can do anything you want with your $300M in private funds, including get it as small bills and roll around in it, but the p-corp funds need to be much more closely monitored. That alone, even without a board, would be big.