Elon Musk's DOGE team may need a crash course in COBOL
fastcompany.comThe problem isn't that they don't know COBOL. The problem is they're not asking questions about things they don't understand. That's very dangerous in legacy code environments.
Maybe this is a dumb question, but why would they have to make changes to cobol at all if there stated goal is to reduce spending? Wouldn’t it be more like figure out the outflows and just auditing spending and then turning some off?
they're operating under a department created by Obama to improve software, so while the department musk is "running" has wildly different goals than the original department did, I'm thinking dorking with software is the ONLY leverage they have. Unfortunately for the country, we use lots of software lol insert sad trombone sound here
If it were an honest attempt to reduce spending, wouldn't it happen in congress?
So many Elon threads, some with conservative defenders, no mention of Chesterton’s fence by anyone, when it’s so apt.
More importantly, Chesterton’s fence is a conservative idea. And it’s an important idea. I am not a conservative, but I am not a total idiot so I can appreciate the wisdom of _not tearing down the fence without understanding why it’s put up there._
If you remove everyone who says it cant be done from the room you are left with people who all think it is possible.
(hahaha)
That's pretty much what the guy in charge of the Titan sub said.
Well, they DID dive...
"The use of COBOL cripples the mind; its teaching should, therefore, be regarded as criminal offense" Edsger Dijkstra, 1975.
Musk runs companies where software mistakes can kill people. I would give him the benefit of the doubt when it comes to software.
Musk runs a company that makes a product called "Full Self Driving" that cannot safely or legally be used for full self driving, and has therefore injured hundreds and killed north of 50 people. Under no circumstances should be be given the benefit of the doubt with anyone's safety, finances, or wellbeing.
This is a kind of weird rationalization where the conclusion leads the interpretation of the facts.
It’s hard to get away from the fact that space x and Tesla deal with critical systems. And by most accounts, they’ve pushed things forward
Yes, and they studied and tested years to get it right.
This is pre-alpha test in productive systems
The nice thing about software is you don’t need the benefit of the doubt. There is no doubt. It either runs or doesn’t. Your website is either hacked or isn’t.
Musk long ago lost the right to be given the benefit of the doubt on anything. The man lies constantly, has the temperament and self-control of small child, an ego the size of Jupiter, is reportedly a ketamine addict, and has verifiably silly things about software design and development. He's completely self-serving. Everything he says is for his own benefit. If you view his statements through that lens you'll quickly have a better understanding of what is happening. I have a similar lens for Trump: The opposite of what he says is closer to the truth. It's incredibly reliable.
> has verifiably silly things about software design and development
Even if you're willing to ignore everything else you said, you'd think a group like Hacker News commenters would recognize how obviously out of touch Musk is with anything software related.
Musk runs companies where software mistakes killed people but he doesn't change his behavior.
I tested it and it turns out LLMs can follow commands such as "Port the following Python code to COBOL", although it's certainly harder to validate the output is correct.
No need to check. It isn't.
You're porting in the WRONG direction. The existing COBOL code is undoubtedly using system utilities such as database management systems, etc. on the mainframe computers. And I doubt that you've written Python code that smoothly integrates into that system environment and just needs to be ported to COBOL. And I doubt that the mini-Musks can do so. (I'm not including the massive application-specific knowledge that you and they would or should need to grasp before attempting to insert code into the system.)
You don't understand. Your COBOL code may be 99.99999% right and still cause millions of dollars to vanish :poof: in a flash.
If the DOGE team have this bright idea of getting an LLM to mess with the COBOL on mainframes there is going to be widespread chaos beyond human imagination.
Or, payments during an emergency. Say, avian flu.
(to be clear this is most definitely not an endorsement of this approach, but you know DOGE is going to try it anyways)
It's useless to port the code if you don't understand the data.