DOGE's AI use in government doesn't add up | Hacker News Reader

DOGE's AI use in government doesn't add up

washingtonpost.com

72 pointsJtsummers15 comments

People that have never interacted with a legacy system don't understand how the legacy system works and think they know how to fix everything.

If you haven't had to deal with people like this before you just didn't write software for long enough. Pretty common in junior devs or people that believe clean code.

I also believe there is a difference of expectations. The legacy system is often written to work with all edge cases that pop up, and in government software also to conform to the law.

If you don't know the requirements you are bound to see something that is inefficient.

That has been every single legacy system i've seen and the last time i worked with the government was 20 years ago.

Once something runs for long enough, with millions of people using it, you will acquire corner cases and if you try to rebuild it from scratch you'll just introduce all the corner cases again, but now one incident at a time.

Not only edge cases but a whole lot of backwards compatibility for both data and other even older systems.

At a previous job there was a huge incident because the order of events was changed, one was arriving a couple minutes before now and it completely broke a considerable amount of downstream consumers.

I once saw how hairy and massive the SMTP implementation was at one of my first jobs and figured I could write my own much simpler drop in replacement. The first word of SMTP is Simple, after all. Why would they write all that extra code that my weekendware PoC didn't need?

The hubris of the junior developer hahaha, its still great though, that's what educates you that software is more than a couple lines of code jumbled together, i think everyone should go through something like that at leas once.

With a certain (class of) people, their MO is to try a lot of things, break a lot of things, and learn from what they broke.

Twitter was a lot of the same (kind of thing). The faster they can break things, the faster they will learn.

And if you are working for such people, it doesn’t help to already know what is going to happen. Blindly fire on all cylinders the commands they issue. Let the fall out occur. Then share the fallout with them. Allow them to learn and get better. They are one-man armies. And you are the high school coach of a proud obnoxious “all star” player.

Background: working in this environment for five years.

Its completely fine to break twitter, if you break social security you'll be breaking people's lives.

One could argue that that is also completely fine for those in charge.

>With a certain (class of) people, their MO is to try a lot of things, break a lot of things, and learn from what they broke.<

...you mean particle physicists

It’s just malice. Hanlon’s razor turns out to be wildly unsafe whenever it becomes common knowledge that Hanlon’s razor is in use.

The page has two titles, I submitted with the shorter one. The other title is:

> No, AI can’t decide whom to fire — or many other DOGE goals

https://wapo.st/3XO0VaL

https://archive.ph/fVmOn

ai everywhere

Reply on news.ycombinator.com