I don't care about platform specific stuff. I'm talking about C which is actually intended to be portable. Nothing written with portability in mind in the past ~decade is going to be doing this.
C is not node.js. C exists for 50 years and is expected to have stable API. In scientific circles it's not unusual to compile c and f77 libraries built in the 70's, 80's.
BLAS, gemv, GEMM, SGEMM libraries are from 1979, 1984, 1989. You may have seen these words scroll by when compiling modern 2025 CUDA :)
In AVR or other MPU-less architecture you can literally modify the string literal memory without triggering a crash.
Why? Because there is no memory protection ("rodata") at all.
And such microprocessors are still in use today, so it's a bit too far fetched to say "really old code."
It's UB, sure, but how many embedded programmers actually care? The OP's proposal is trying to change the type system so that this UB becomes much less likely to trigger in practice.
> Sometimes, this sort of “misfire” of the “known answer” circuit happens naturally, without us intervening, resulting in a hallucination. In our paper, we show that such misfires can occur when Claude recognizes a name but doesn't know anything else about that person. In cases like this, the “known entity” feature might still activate, and then suppress the default "don't know" feature—in this case incorrectly. Once the model has decided that it needs to answer the question, it proceeds to confabulate: to generate a plausible—but unfortunately untrue—response.
Confabulation means generating false memories without intent to deceive, which is what LLMs do. They can't hallucinate because they don't perceive. 'Hallucination' caught on, but it's more metaphor than precision.
It is possible, although not confirmed, that the block was motivated by an ongoing malware threat that involves this particular threat, but take this with a grain of salt.
