good luck, everyone who hates networking and doesnt want to learn it and thinks they can ask ai about it and survive

[viss@mastodon.social]

good luck, everyone who hates networking and doesnt want to learn it and thinks they can ask ai about it and survive

[kajer@infosec.exchange]

@Viss

lolollolol

interface went down, better restart BGP

[ai6yr@m.ai6yr.org]

@Viss Autocomplete trained on bad data = bad advice.

My theory is there are a lot of reddit posts that include the BAD examples (the non working ones), and that training corpus goes into the network and WHAM.

So if there are posts from 10 people about "THESE COMMANDS DON'T WORK, HERE'S WHAT I TYPED" the 10 bad commands get reinforced into the neural net... lol.

[kevinmirsky@infosec.exchange]

@Viss "You're absolutely right, my mistake. [..]"

[paul_ipv6@infosec.exchange]

@Viss

they'll find out the hard way that if you keep hating learning networking, networking will hate you right back.

[jamoquanty@mastodon.social]

@Viss I've found llm's pretty good for networking stuff ime.

Much like any decent tool: foot shooting is a feature not a bug.

[pauliehedron@infosec.exchange]

@ai6yr @Viss

I deal with this on MacOS all the time. LLMs can't write correct zsh because most of the corpus is linux bash, but their bash is bad because the MacOS userland is BSD so all the cli tool calls have wrong options. So I have to tell it to use POSIX sh and no awk to have stuff work. 🫠

[quatermasstools@infosec.exchange]

@paul_ipv6 @Viss and if they do relent and learn networking they’ll understand why the network hates them