iiinova — where your digital self keeps evolving

i think claude finished the task i set it and then it immediately killed its own box by running the big test suite instead of the small test suite. i guess i know who's skill is issued now

000

like butterfree is a butterfly, beedrill is a bee. a buncha pokemon are birds and whatnot. but i don't think i would have got pikachu=mouse

i don't think it would have ever occurred to me that pikachu was meant to be a mouse if i wasn't told that

got claude to generate some nice documentation somehow. it turns out that "LLM voice" was actually just a skill issue on everybody else's part

Jenna JJ Q. @tomatofacts3.bsky.social · 08/09/2026

Reposted by Jenna JJ Q.

cant believe openai has proved the polynomial hierarchy collapses to the 37th level

002

Jenna JJ Q. @tomatofacts3.bsky.social · 06/10/2026

i used to meet people who use my software and they'd have interesting feedback about it & that was cool. today i meet people who have agents use my software and they don't have anything interesting to say about it

1121

Jenna JJ Q. @tomatofacts3.bsky.social · 03/10/2026

update: tried this on my shower at home, and it does work, though not quite as well

for user-defined types, you can implement the Deref trait, but Deref is safe, and dereferencing a NonNull would be unsafe. And there's no UnsafeDeref or equivalent (maybe there should be)

000

I think it's because there's not really any good way to implement it, at least not without a new language feature or by making NonNull a "special" type recognized by the compiler

100

i was invited to submit my paper to an agriculture journal and i suspect it's purely because the name of one of my research projects is plant-themed

in my project I genuinely have test cases *just* to test the variance + various marker traits of my unsafe types, just because it's so difficult to otherwise know that i managed to get it all correct

040

even though Box<T> obviously mutates through its pointer. In general, const/mut ptrs have nothing to do with constness or mutness, the only reason to pick one or the other is to get the right variance

The entire logic seems to be that &T is covariant and &mut T is invariant; and *const T corresponds to &T and *mut T corresponds to &mut T so they have the same variances. But this leads to situations like Box<T> using *const T under the hood since it needs to be covariant

141

There's no principled reason for *const T to be covariant and *mut T to be invariant; neither type carries any type invariant that T exists in memory. They could be *bivariant* and it'd be fine!

It's like the Send/Sync situation but worse, since there isn't even an unsafe way to expand the variance. The ONLY way to configure the variance for your type is in selecting the types of the fields, so you have to play type tetris to get the desired constraints, possibly with PhantomData

140

do you want to hear another rant

sorry to get off topic a bit. BUT. It is in fact idiomatic to unsafe impl Send a struct that has pointers in it, if that struct is meant to be sent. (This is how Arc is implemented for example.)

That it's not easy to restrict the Send/Syncness of a type is also a design flaw.

That is, by writing 'unsafe', I'm actually opting out of some of the restrictions that Rust applies to my code, many of which I was probably actually relying on. Thus making me write 'unsafe' to opt out of checks when it's fundamentally unnecessary to begin with is bad design

020

A changes and suddenly A isn't Send anymore. Normally I would rely on the auto trait system to catch any problems. But X is still going to be Send.

the concept of 'unsafe' loses its value if it's applied to too many things that don't need it. imagine for example `struct X { a: A, ptr: *mut u64 }` or something. Now suppose I want to Send it. Oh no, the pointer prevents it from being Send, now I have to unsafely mark X as Send. But then suppose

passing the pointer to another thread be treated differently, when that operation by itself has no safety requirements?

<sorry for the rant you, but hit one of my pet peeves about Rust> I don't really believe in something being "more unsafe". Things are either unsafe or they aren't. There are an infinite number of things you can do with a pointer that makes it difficult to uphold their safety conditions. Why should

actually my Hot Take is that raw pointers should be unconditionally Send

220

IMO it's fine to make it Send as long as it's still the case that any problematic behavior requires you to do an unsafe operation

mutating through the pointers is unsafe, though? (I'm assuming you mean raw pointers)

210

Jenna JJ Q. @tomatofacts3.bsky.social · 01/10/2026

could you do 'unsafe impl Send for BlockList'?

210

it's genuinely impressive how many people think we need bespoke pdf reader web apps

Jenna JJ Q. @tomatofacts3.bsky.social · 30/09/2026

It's not just a test framework—it's a means to keep the codebase honest. This test harness sets the thread count to 4: that's how we know it will run fast. Without this, the test harness would have to run without parallelism.

090

Jenna JJ Q. @tomatofacts3.bsky.social · 30/09/2026

i wonder if this works on all tubs or just this hotel's

Jenna JJ Q. @tomatofacts3.bsky.social · 30/09/2026

i accidentally discovered if i turn on the shower, put my head against the bathtub, and put my fingers in my ears, it sounds like there is running water in my head

121

Jenna JJ Q. @tomatofacts3.bsky.social · 29/09/2026

"i don't need these arguments because -" no, you do!!! that's why they're there!!!

040

Jenna JJ Q. @tomatofacts3.bsky.social · 29/09/2026

ive been frustrated lately cause of agents that think they know better than the spec

230

Jenna JJ Q. @tomatofacts3.bsky.social · 28/09/2026

there should be prediction markets for whether a flight will be cancelled

000

Jenna JJ Q. @tomatofacts3.bsky.social · 27/09/2026

they look too warm and friendly

Jenna JJ Q. @tomatofacts3.bsky.social · 26/09/2026

the founding fathers were wise to put washington dc on the opposite side of the country from hollywood. we should have kept them apart

000

Jenna JJ Q. @tomatofacts3.bsky.social · 26/09/2026

i feel like i'm explaining the same things over and over again. the corpo keeps finding new places to pull out gaggles of stakeholders who need basic information

Jenna JJ Q. @tomatofacts3.bsky.social · 26/09/2026

i think twilight but it's cold mermaid girls instead of sparkly vampire guys could have the sauce

Jenna JJ Q. @tomatofacts3.bsky.social · 25/09/2026

to be clear, the PR was mine, the review was AI-written

Jenna JJ Q. @tomatofacts3.bsky.social · 25/09/2026

today i got a "code review" from a coworker that was like 1000 words of LLM slop that he clearly hadn't read. and it ALL sounded like this. i am going to kill myself

140

Jenna JJ Q. @tomatofacts3.bsky.social · 24/09/2026

always in writing you need to consider the mindset of your audience ... llm docs always assume a particularly weird mindset going in

030

Jenna JJ Q. @tomatofacts3.bsky.social · 24/09/2026

look, it is sometimes appropriate to document the "why not?" but the correct form is something like: "It may be tempting to do Y, because <reason>, but this isn't a good idea because <reason>"

Jenna JJ Q. @tomatofacts3.bsky.social · 24/09/2026

llms love to write like "X is the point: don't Y because Z" where X is an incidental design detail, and Y is an alternate design that nobody would ever wonder about or be concerned with

230

Jenna JJ Q. @tomatofacts3.bsky.social · 23/09/2026

that's why they require a higher weirdness index!

Jenna JJ Q. @tomatofacts3.bsky.social · 22/09/2026

the problem with this definition is that I don't really feel that "but" is commutative