octoberfranklin

Also "FooCo raises $X zillion round" doesn't mean they got $X zillion deposited into their bank account. There are always several milestones along the way to unlock each tranche. But the press releases never mention that.

Just find the charge, click the "I want to cancel the service that caused this charge". The merchant has to either cancel or reply with a message clearly explaining why they can't do that (i.e. you agreed to a 12-month minimum).

Whoever is collecting the money has all the information needed to make the cancellation request. No website logins, passwords, self-addressed stamped envelope, etc.

Apple has done this for subscriptions that collect money through the iTunes store. It's awesome.

AmEx should do this because unlike all the other cards they control the whole stack.

Pinky-swear!

you shouldn't trust the third party AI provider either

I trust competing American AI labs to not collude with each other, so I can spread my work across them and nobody sees everything.

I trust Chinese AI providers to not work with the NSA.

And of course there's always tinfoil.ai's attested model engines... probably the only non-evil use of Trusted Computing I've ever seen.

But Cloudflare is different. They're the single point of access for everything. Which is precisely why they attract irresistable government pressure. Trusting them is like painting a target on your back.

And I think OP's point, more succinctly, is that Valen programs will be much more difficult to parallelize than Rust programs.

It's shockingly easy to make a single-threaded Rust program use all the cores on a machine by slapping in Rayon wherever you have a Vec. Because Rust forces you to do the hard work of proving shared^mutable before getting a single-threaded program running.

The ecosystem-wide consequence of this is that pretty much every compute-intensive program written in Rust (that doesn't rely on non-Rust libraries for compute-intensive stuff) is automatically multicore. This is one of the reasons why Rust programmers seek out Rust libraries first. Because they know they won't get the unpleasant surprise of putting in a lot of work to adopt a library and then get burned when they find out it will only use a single core.

The problem with this is that it imposes a cost on abstraction. Zero-cost abstraction is one of the most fundamental design principles behind both C++ and Rust.

A field-path into a struct/enum fundamentally depends on its concrete implementation. If you hide the fields of a struct and use getters/setters, you break the ability to talk about the "paths modified" by a function which uses getters/setters.

Aside: it really annoyed me that I had to read halfway through this article to find the first attempt at defining "group borrowing". The whole first half of the article is basically fluff.

The model engine provider might be ZDR, but the service as a whole isn't.

So yeah, among the few physicists who understand enough set theory to comment on this, most of them are totally fine with admitting that our reality is not based on set-theoretic real numbers.

For mathematicians, sets are the base reality and everything else is constructed in terms of those. For physicists experimental validation is the base reality, and if it disagrees with set theory that's okay with them; it just means that the set-theoretic real numbers aren't the correct real numbers for doing physics. They're very pragmatic when it comes to mathematical foundations, and that's probably the right approach to take for doing physics.

There have been a few attempts to formalize real numbers that more closely match our observed reality; look into measurable sets. Most of these require a large cardinal, so they go beyond ZFC.

... and she complied. Looks like he chose his accountant poorly.

There are also a bunch of other stinkers, like building a Turing machine out of Navier-Stokes fluids -- except that it only works if you can encode literally infinite amounts of data in the relative positions of two particles. I.e. assuming physics is based on set-theoretic real numbers, something we've known is wildly false for over a century: https://en.wikipedia.org/wiki/Banach-Tarski_paradox

Just like vuln reporting, the AI industry has put zero effort into triage here, and the models are really good at making their findings sound more important than they really are.

The cynic in me suspects this is a smokescreen for the Navier-Stokes tokenstream plagarism fiasco.

There was no economically justifiable reason for me to learn about compilers, operating systems, and even CMOS VLSI. But I couldn't be satisfied writing software without knowing how those things worked. I taped out a chip in which all the polygons drawn by deterministic software I'd eithe written or understood. No place+route. It was the 2000s version of non-AI coding in the 2030s -- standard cell placement is essentially gradient descent.

Some people just have to know how it works.

I would be astonished if they weren't already doing this.

That is totally disappointing.

I have eight or nine custom extensions at varying levels of maturity, some nearing the point where sharing them becomes useful.

One is a simple but obvious Eoka extension for Pi.

I modified my terminal (Alacritty) to support the Kitty image protocol (it already spoke sixel; Pi understandably uses the superior Kitty protocol). And a toolcall that lets the LLM put an image in the transcript (upstream Pi lets only image-modality LLMs do this). "Deepseek, do XYZ in a headful Chromium using Xvfb and show me a screenshot before each step".

Once you see the immense power you get from being able to modify both ends of the terminal connection (and having LLMs write the tests and tedious parts for you) it's addictive. I've added a custom terminal extension for negative-row-number cursor positioning, and support for it on both sides of the connection (Pi and Alacritty). End result: all the performance of native scrollback with no compromises -- Pi can modify rows above the top of the visible part of the screen, like collapsing/expanding toolcalls, without forcing a full-terminal refresh.

Also: click-to-expand/collapse toolcalls/replies individually. Really helpful.

Disappointed that Pi is abandoning the minimality philosophy.

Pi's selling point used to be: no MCP, native scrollback.

Now Pi is a fullscreen TUI with built-in MCP.

Man, first MCP then this? The main reason I use Pi is because it didn't try to reinvent my terminal's buttery-smooth native scrolling -- something no terminal client can ever match.

Did you all get acquired by private equity or something? The enshittification is coming at us fast.

They are.

The question is whether we want mathematics to be dominated by whoever has the most capital. This is perilous because mathematics doesn't produce profits like commerce does. The AI labs might end up killing something off that they are unable to sustain a replacement for.