Thanks for sharing this! I really think that when people see LLM failures and say that such failures demonstrate how fundamentally different LLMs are from human cognition, they tend to overlook how humans actually do exhibit remarkably similar failures modes. Obviously dementia isn't really analogous to generating text while lacking the ability to "see" a rendering based on that text. But it's still pretty interesting that whatever feedback loops did get corrupted in these patients led to such a variety of failure modes.

As an example of what I'm talking about, I appreciated and generally agreed with this recent Octomind post, but I disagree with the list of problems that "wouldn’t trip up a human dev"; these are all things I've seen real humans do, or could imagine a human doing.

String escaping sucks in bash and other posix-style shells too, though.

Most of those comments are actually just random people arguing about the merits of the experiment, not continued discussion with the bot.

Also, the bot is supposed to be able to run builds to verify its work, but is currently prevented from doing so by a firewall rule they're trying to fix, so its feedback is limited to what the comments provide. Humans wouldn't do great in that scenario either. (Not to say the AI is doing "great" here, just that we're not actually seeing the best-case scenario yet.)

It's a "stream manipulator" function that not only generates a new line, it also flushes the stream.

Yes, the dialog was changed, as part of this linked issue (and maybe again after that; this whole incident is very old). After reading some of the comments on that issue, I agree with the reasoning with some of the commenters that it would be less surprising for that menu option to behave like git reset --hard and not delete tracked files.

If you look at the proposal, this is specifically "static reflection", i.e. compile-time reflection. So it doesn't actually have any of the downsides you mention, as far as I can tell.

There is nothing special about C.

I wish that were true, but isn't it somewhat wishful thinking? Even an assembly-language Lisp would require an operating system in order to build a functioning compiler, wouldn't it? And operating system APIs are in C.

Edit: more importantly, as the post explains, the special thing about C is the existence of TinyCC.

In this context, stabilization refers to the adoption growth curve flattening out.

I'm not totally clear on why signals are used here in the first place. Arguably most C code doesn't "need" to use signals in complex ways, either.

I've met people with C++ Stockholm Syndrome, and I think their trajectory is different. There's no asymptotic approach toward zero; their appreciation just grows or stays steady, even decades into their career.

I'm pretty sure that when programmers and other techies call themselves "hackers", they don't mean in the security-breaching sense. It means that you can "hack together" something.

One problem is that the build isn't easily reproducible: there are a few comments in that issue thread from someone trying to reproduce it and failing.