jeffparsons

> just stops shy of asking Australian tech companies, like Fastmail, to build backdoors into their products so that the government can "legally access" data from them

It stops just short of saying that you must do thispreemptively, but is pretty clear that you must do it if they ask you to.

Basically the same phrasing was widespread before LLMs were invented. So I don't think it's specifically motivated by that.

I don't have any special certification or arrangement with Anthropic; this is vanilla Opus 4.x via Claude Code.

It may be worth noting that Claude can and will (if it believes you own the code, at least) produce PoC exploits for exploitable bugs that it finds.

My only source for this is personal experience, and no, I can't share any evidence of it.

I've regularly heard something similar said of consulting work, too. Many people new to the game worry about charging too much, because if a client is paying more then surely the pressure will be higher. Instead they end up experiencing the opposite: charging a higher rate tends to get them a better kind of client.

I'm not sure what the exact lesson is here. Something about stingy people not being nice to work with, perhaps?

Amazon, for example, charges us for cloud resources and then charges us again (handsomely) for the privilege of submitting bug reports to them. And then sometimes, even with a clear, deterministic repro for a bug with no plausible workaround (besides "stop using the feature"), where the fix is probably as simple as "pull a fix from upstream open source repo" or "sic Claude on it for 10 minutes", the bug remains open for literally years.

This is very different from "I didn't read the instructions on the screen and now I'm calling support". Both scenarios exist. I have some sympathy for businesses facing the latter, and much less for businesses facing the former.

Claude won't answer questions about what cities you should nuke in what order. The Pentagon wants Claude to answer those sorts of questions for them.

Edit: oops, I misunderstood. This seems to be more about contractual restrictions.

Or in case some folks find the addition of a computer confusing here, if someone sends you a physical letter and they've used correction tape or a black marker to obscure some parts of the letter, and you scratch away the correction tape or hold the letter up to a light source to read what's underneath, have you committed a crime?

I'm not a lawyer, so I don't know what the law has to say about this. But I do have at least a small handful of brain cells to rub together, so I know what the law _should_ say about this.

It's possible without specific hardware acceleration, but murderous for mobile devices.

RDS, Route53, and Elasticache are decent, too. But yes, I've also been bitten badly in the distant past by attempting to rely on their higher-level services. I guess some things don't change.

I wonder if the difference is stuff they dogfood versus stuff they don't?

Additionally, what has been the correct choice five years in a row might be catastrophically wrong in the sixth year. We need some randomness injected into our behaviour so that some people are always making "suboptimal" choices, to stop everyone from crowding into one local maximum and then getting swept away when the rare but inevitable flood comes along.

And with a little work you can even use them to map ranges of keys to values in a way that's reminiscent of interval trees — e.g. https://crates.io/crates/rangemap. (Disclosure: that's my crate.)

This email is from a Debian maintainer, about Debian introducing a new hard dependency on Rust. It's not some random Rust advocate telling Debian folks that they should use Rust against their will.

Yes there are absolutely some obnoxious "you should rewrite this in Rust" folks out there, but this is not a case of that.

Good question. I'm not primarily blaming Discord or the other company for this (even though they both obviously share some responsibility, too) — I'm blaming government/legislators. I'm arguing that the government agencies/departments that own the relevant forms of ID should have been required to develop the capability to facilitate this sort of secure ID verification _years_ ago. Instead policy makers ignored reality and rushed through this legislative hatchet job... and here we are yet again. As anybody who's been awake during the last few decades could have predicted.

Tangent: I've regularly been required to provide copies of my ID to all kinds of businesses simply to function in society — i.e. in practice there is no realistic option to opt out. Want to rent a house? X points of ID. Want a phone? X points of ID. Pretty much every real estate agency in town has copies of at least my driver licence. And they in turn share my details with tenant database companies, credit reporting agencies and so on. Do you think many of these businesses have good data handling practices? Of course they don't. And so all my details are available for purchase in bulk data sets on the dark web, and get refreshed by new data breaches every few years. And yet government still treats it as somehow unexpected each time this happens, or wags its finger and bemoans those naughty criminals, instead of developing any kind of policy that would start to address the underlying issue... which is that our personal details are spread so far and wide in the first place.

The third party company shouldn't ever need to see the IDs, either. Same issue.

There is no reason for a company like Discord to ever see the ID. The owner of each relevant form of ID — usually a government agency/department — should provide an attestation service, such that users prove their identity to the agency and the agency tells the company "yes, this user is who they say they are".

It's not that hard. Legislators around the world are consistently dropping the ball on this.

Don't get confused here: they didn't decide that their policy change was wrong — they just didn't expect quite as much backlash.

Make your purchasing decisions accordingly.

I think the person you were replying to might have intended sarcasm.

My Android phone prevents me from taking screenshots if an app author doesn't want me to.

My Android phone prevents me from recording phone calls at the request of my carrier, even though it's totally legal for me to do so in my jurisdiction.

I'm not loving where this is all going.

In Australia: Centrelink, Service Victoria, Medicare, myID (formerly MyGovID), ATO (tax) — for a start.

There's no great reason for these to be Android/Apple specific. I'm just offering examples as requested.

I took a swing at something vaguely similar a long time ago now: https://github.com/jeffparsons/planetkit

My approach was to build a hex grid on a geodesic sphere. It's a very different trade-off.

I didn't specify the challenge clearly. I meant to allow blocks in any orientation, as long as they would be stable.

So you can, for example, have blocks sloping down from the edge of the table by sandwiching one end of them between two other blocks with enough vertical distance between them, and enough weight on top.

This would be easily solved for customers who care about it by allowing you to pay a one-off fee to reserve the name for ~100 years.

Or they could just absorb that.

Any idea why it works that way? Have they offered an explanation?

I'm a Fastmail customer but I've never noticed this because I use my own domain.

That's basically the direction I was going in my head. I just remembered we have a bunch of Kapla blocks in the house, so I may be able to do this "IRL"!

> My goal here is to develop an intuitive sense of comfort with the behaviors of these stacks. If I succeed, you will not just understand that the physics allows the stacks to be stable, but you will feel that it is proper and just.

I love this kind of writing. It feels like the author is excited to bring me along on a journey — not to show off how smart they are. In this way it reminds me of Turing's original paper that introduced his "computing machine". It presents a fantastically deep topic in a way that is not just remarkably accessible but also conversational and _friendly_.

I wonder why so little modern academic writing is like this. Maybe people are afraid it won't seem adequately professional unless their writing is sterile?

How about this one:

Assume an arbitrarily high coefficient of friction between all surfaces. Can you stack the blocks on the table such that at least one block is wholly below the top of the table?

I think I have an answer to this, but I've only worked it through in my head, so there's a good chance I'm wrong!

I've often thought that I'd prefer it to check by default in release builds, too, but I understand that comes with a performance penalty that a lot of folks aren't happy with.

I assume this implies that common processor architectures (x86_64, aarch64) lack trap-on-overflow variants of their integer arithmetic instructions? If the explanation really is that simple, it's pretty disappointing.

Why is Rust a bad example? Of the code bases I've tried Claude on so far, it's done the best job with the Rust ones. I guess having all the type signatures there and meaningful feedback from the compiler help to steer it in the right direction.

Has your experience been different?

I don't have a source for this, so take it with a huge grain of salt... but for some reason I have a memory of someone telling me that the older versions of Word saved and loaded documents by writing the bytes of in-memory data structures directly to files on disk, with not much in the way of marshalling or validation in the middle. Because it was fast, or something. You can imagine the kind of edge cases and oopsies that might result.

The new versions at least serialise to some kind of monstrous XML representation of Word's internal state, so while it's not going to win any awards for world's most elegant document format, it should be slightly harder to corrupt in subtle ways.

Another reason to use LaTeX for papers back in the day was that Microsoft Word would routinely corrupt large documents in terrifying ways. Sometimes the root of the corruption existed in the document somehow long before any of it was visible, so even recovering from an old backup would just lead to the problem repeating. I recall the only way to properly "recover" an old backup was to copy it all via plain text (e.g. Notepad), and then back into a brand new Word document.

This is all to say, if you're working on a theis or even a moderately large assignment, working in Word was not good for the nerves.

Looking back, I probably should have just worked in plain text and then worried about formatting only at the very end, but ummm, yes, I guess another hapless victim did indeed fall into LaTeX's trap. :)