The Pentagon’s Budget Is So Bloated That It Needs an AI Program to Navigate It | Hacker News Reader

The Pentagon’s Budget Is So Bloated That It Needs an AI Program to Navigate It

theintercept.com

36 pointsjram9308 comments

I have yet to jump on the "LLM" bandwagon and think ML should be used sparingly but this is one thing I believe is a cause for applause.

Humans like money and get bought off on such endeavors but ML doen't. Humans get bored and lazy but ML doesn't. If this isn't a good application of it, I don't know what is.

A pattern I have seen in corporate bureaucracy is the heavy cost of laziness by middle management that results in waste, misaligned incentives and ridiculous red-tape. All because mid-level bureaucrats can't be bothered to make granular decisions or enact case-by-case policies because they fear all that work will distract from all the lunches, meetings and forced team outings. I wouldn't be surprised if DoD has a similar bureaucrat problem. ML based software is the perfect bureaucrat. However, still having a bribable human in the loop is always nice.

These articles write themselves i guess. If it weren't a negative article about the dod using ai, it would be a negative article about why isnt the dod using ai.

What about an ai generated article about the dod?

From the ai.mil blog post linked in the article:

>GAMECHANGER is an AI-enabled policy Natural Language Process (NLP) search capability and centralized document repository for the Department of Defense (DoD)

I am all for uncritically bashing on the burning money pit known as the Department of Defense, but from this description I do not get the sense that this is some cutting edge LLM type AI tool, nor does its existence seem controversial. If anything, I'm surprised to learn that the DoD did not previously have a searchable, centralized document repository prior to 2018.

Am I missing something?

It kinda sounds like yet another ChatGPT wrapper, but for the DoD, maybe with some fine-tuning to the database specifically. Microsoft has a cozy relationship with the Pentagon and IIRC they just switched all their cloud infra to Azure so it's not unreasonable to think it's LLM-based.

I'm not as deep in the NLP space as some people so maybe neural text search really did improve a lot the past few years but when I was more involved it was not really as successful as it should have been given all the hype at the time. (Probably it's still not, but may be a marginal improvement despite it being an obvious cash grab for a private contractor somewhere.)

TBH I mostly responded because I am really curious about your username: what inspired it? TOoCitBotBM (if you can parse that) was a foundational book for me in my mental model of the world I live in.

Sorry to burst your bubble here but there's no special meaning to my username. Whenever I created this account I was trying to think of a psudonymous name and I just happened to have The Origins of Consciousness... sitting on my desk. I should probably change usernames and also finish reading that book someday, haha.

I understand the benefit of this solution, but it seems to me that this will enable more of the problem. Now we have AI to figure this out, so let’s go wild on duck taping more policies. Let’s even add contradictions and it will just give you an answer when you ask for it. Seems they’re afraid of making the tough call of simplifying it.

Why say bloated when you could say complicated?

Reply on news.ycombinator.com