> People should not hear loud music when driving - max is normal speaking voice level.

which feels like a more than acceptable constraint to me.

People can shout "domestic terror" all they like, but if it's not true, it's not true.

It is (rightly) difficult to get hold of one uninvolved child, for safeguarding reasons, so it would be better to run it as a school (or interschool) competition, where multiple children may participate. For fairness, you may also provide multiple LLM participants (however you define that). The winner of the contest, as determined by the judge, would then determine the winner of the bet ­– unless the winning child had been trained, in which case we would fall back to the next-highest-ranked participant. The number of LLM candidates would be equal to the number of eligible children.

However, I don't see a good way to allow each child to pick a programming language and task, without leaving the competition results incomparable. So perhaps each child should be paired with an LLM, and the judge should determine which submission from each pair is better? But then if I only need one victory (to support my claim), this is clearly unfair. So each pair should be tested enough to determine whether they're consistently better than the LLM… but then we are demanding a lot of the child participants, for no real benefit to them.

If we can agree on a workable protocol, I can try to pull some strings and see if we can make this happen. I could use the money.

I don't care how many terms you add to your Taylor series: your polynomial approximation of a sine wave is never going to be suitable for additive speech synthesis. Likewise, I don't care how good your predictive-text transformer model gets at instrumental NLP subtasks: it will never be a good programmer (except as far as it's a plagiarist). Just look at the Claude Code source code: if anyone's an expert in agentic AI development, it's the Claude people, and yet the codebase is utterly unmaintainable dogshit that shouldn't work and, on further inspection, doesn't work.

That's not to say that no computer program can write computer programs, but this computer program is well into the realm of diminishing returns.

> There's a common rebuttal to this, and I hear it constantly. "Just wait," people say. "In a few months, in a year, the models will be better. They won't hallucinate. They won't fake plots. The problems you're describing are temporary." I've been hearing "just wait" since 2023.

We're not trending towards superintelligence with these AIs. We're trending towards (and, in fact, have already reached) superintelligence with computers in general, but LLM agents are among the least capable known algorithms for the majority of tasks we get them to do. The problem, as it usually is, is that most people don't have access to the fruits of obscure research projects.

Untrained children write better code than the most sophisticated LLMs, without even noticing they're doing anything special.

They're also still tonnes of metal hurtling along the streets of a city shared by pedestrians, which is inherently dangerous. (Less so than a bus, but there are also more cars than buses: you'd have to check the statistics to see how that evens out.) As for actually damaging the road (producing road dust, potholes, etc, requiring a resurface that off-gases for weeks afterwards): cars damage the road more than bikes, though that's not significant compared to lorries, since the wear is something ludicrous like the fourth power of the weight-per-axle.

And yet, you can write pure-functional thunked streams in Python (and have the static type-checker enforce strong type checking), and high-level duck-typed OO with runtime polymorphism in Haskell.

The hardest part is getting a proper sum type going in Python, but ducktyping comes to the rescue. You can write `MyType = ConstructA | ConstructB | ConstructC` where each ConstructX type has a field like `discriminant: Literal[MyTypeDiscrim.A]`, but that's messy. (Technically, you can use the type itself as a discriminant, but that means you have to worry about subclasses; you can fix that by introducing an invariance constraint, or by forbidding subclasses….) It shouldn't be too hard to write a library to deal with this nicely, but I haven't found one. (https://pypi.org/project/pydantic-discriminated/ isn't quite the same thing, and its comparison table claims that everything else is worse.)

The best LLMs can manage is "what's statistically-plausible behaviour for descriptions of humans in the corpus", which is not the same thing at all. Sometimes, I imagine, that might be more useful; but for programming (where, assuming you're not reinventing wheels or scrimping on your research, you're often encountering situations that nobody has encountered before), an alien mind's extrapolation of statistically-plausible human behaviour observations is not useful. (I'm using "alien mind" metaphorically, since LLMs do not appear particularly mind-like to me.)

> It was never just about the candy. It was about being together.

And then, a subsequent paragraph ends:

> But the ritual remained.

I'll allow "Grandpa’s chocolate was something else entirely. It was sacred." because (while it fits the pattern) it actually works; but we see the same pattern again later:

> It’s sweet. It’s bitter. It’s ours.

I'll give the benefit of the doubt for the overall structure of the narrative, too, because maybe the author's picked up (or, heck, perhaps pioneered) the traditional Facebook Anecdote Genre. I'll even give a pass for the style, because it resembles the writing style of https://www.csmonitor.com/1996/0502/050296.home.home.1.html – if somewhat less skilful. But those parts I've identified as least skilful are also those parts I've identified as most AI-like.

Now look at when Nancy Intrator contributed to the Christian Science Monitor: 1994–1997. Sure, it's plausible that she's contributing again after a long haitus (and the facts of the story back this up)… but we've got a lot of "plausible"s adding up.

You mentioned the en-dashes (which I don't treat as an AI indicator; I use them myself, even!). Nancy's 90s writing did use dashes, but much less frequently: https://www.csmonitor.com/1994/1228/28172.html has 4 dashes in a much longer piece (all spaced ASCII, but we can chalk that difference up to software), whereas TFA has 11! (https://www.csmonitor.com/1995/0329/29161.html has no dashes at all, but it's a slightly different genre of anecdote, so I'll ignore that as an outlier.)

If I had to guess, I would say that there was a shorter draft written by the real Nancy Intrator (or someone pretending to be her, but probably the real Nancy), which was then expanded by a generative AI system. The additional research I've done for this comment has not changed my guess.

When there's a crisis, it's still worth checking in to see what the LessWrongers are saying about it, because it might be very useful, and it's pretty easy to tell: you just check whether it looks like they're doing science, or Rationalism™®, and only investigate further in the rare cases where it's the former.