> The irony here is perhaps missed by many. He didn’t grade homework in the first place. He paid assistants to grade.

Whether he graded it, or his assistants graded it, I’m not really sure how that matters at all to the point.

Strangely, the article never posted what the request actually was.

There usually is a fee for the amount of information, or for the scope of the search, lest they get spammed for low value drive-bys.

If the request was “anything anywhere anytime related to X” they could be expensive (millions is incredulous, but expensive).

If the request was “last 9 months arrest records by X” that is more scoped.

The article never says.

As in, the email was misrouted?

Or simply it falls into no one’s job?

The latter is deliberate. The former, maybe.

In my case it is the latter.

In one example, someone wanted all internal training, presentations, engineering documentation, internal policies, etc related to the use of AI to accomplish a specific goal.

I was temporarily assigned to be a kind of product manager for the project, so I have the best access to all the multiple departments with 100s of staff involved. That said, I have no authority over them. I don't know every related policy, who developed what training, what engineering documents exist in some random share drive.

It meant calling meetings with the project leads to explain the request. Then hassling them to hassle their staff to check all their drives. Reading through all that material to ensure engineering didn't screen cap PII while making a training slide or put security-sensitive details in an engineering document. There was no guidance on what is not releasable.

And we get maybe one or two of these requests a year for different projects. So it wouldn't be a judicious use of taxpayer resources to build a reliable system for answering these requests.

Eh, I'd say it's way more complex than that. State agencies are under budget and staffing constraints just like regular businesses. Every employee tends to have more work than they can finish, and tax payers don't want to increase funding or want to actively lower it.

It’s definitely deliberate. It’s not like this is a natural phenomenon. The only question is who makes the choice and whether it’s a choice specifically to make the FOIA process painful or merely not to give it enough resources and priority.

It's deliberate but not malicious. Like when my kids ask for a new toy. Saying no because we can't afford it is deliberate but not malicious.

State agency resources are finite. I know my small town had to jack up property rates like crazy this year just to make its slim budget (it's the largest source of revenue, and the 2d is sales tax, which is unfortunately in the toilet thanks to several ruinous federal policies causing people to spend less and fewer people to come as tourists).

If we got FOIA requests, lol, it would take years to respond and would be so expensive. We don't even have a city attorney position. It's contracted out to a law firm because it's cheaper.

Some big FOIA request would cost us an insane amount of money through no fault of our own. Our voters don't want to pay higher taxes, so that's the state of things.

> Our voters don't want to pay higher taxes, so that's the state of things.

So the alternative is to act against the law and not answer FOIA requests? You should demand high enough taxes to fulfill your obligations. If voters are unhappy about their height, they should elect people who make different laws the next time.

Often, as with the Eugene police in the source article, the voters who pass a requirement are not the same as the voters who have to approve funding to comply with it.

Where do you think the internet came from?

Neither of those things. Arpanet was military and the web came out of CERN, but they were both low budget side projects.

So…you’re point is that only his judgment says it’s useful, and maybe he shouldn’t be using it and instead rely on…his judgement.

No, my point is LLMs are machines designed to create responses to prompts that pass as plausibly human and intelligent. They are working as designed here. The fact that they get stuff right on occasion doesn't change that what it is at it's core is a randomness generator.

They are designed to create responses that pass.

Randomness generators would fail horribly.

Because...that's what the PHP security team does?

No one needs to use their work, or indeed PHP at all.

But that's why the PHP security team exists.

If it takes 2 months to do fix a massive security issue, it's best if someone else steps up to the plate to fix it.

I don’t think the parent would disagree.

But the PHP team is the one that got the security disclosure.

I saw it in earlier Claude models, but I haven't seen it for a least 7-8 months.

Are you victim blaming the reporter?

If you have a bad process, fix the process.

When you say Claude…you mean the CLI?

That’s a harness.

I mean both the model and the CLI.

That's a harness, yes, but a harness can be more than this.

This gets mentioned on HN at least 5 times a day.

Highly relevant then. HN tends to discuss the things that are either meaningful today or what they are building in their garage that is often meaningful in 5-10 years.