Ian Campbell 🏴23h ago

Okay, okay. I need to devote some time to catching up on genAI capabilities in a professional sense.

Security Researchers & SecOps - what's your favorite use case so far?

Also, what's a lesson you learned the hard way?

***Also - please save the snark. I'm tired, and this is a genuine, if hesitant, ask.

#infosec

Show threadIan Campbell 🏴
One of my professional networks - filled with actual practitioners - is substantially less negative on AI lately. There's spend and craft involved, so it's not a turnkey solution, but a lot of folks in this trust group are finding substantial productivity benefits, rather than hype.
Apr 2 at 2:19pmWeb
Show threadDavid Penfold 23h ago
@neurovagrant Yep, same. Just a pity I had to quit my job because those specific people didn't have a clue.
Show threadViss23h ago
@neurovagrant they've moved on to 'harness engineering' from prompt engineering. i can show you what I've built if you'd like
Show threadIan Campbell 🏴23h ago
@Viss would definitely appreciate any experience you feel like imparting
Show threadViss20h ago
@neurovagrant just let me know when you have some free time today, if youre game. i think my entire day is earmarked to deal with with it all
Show threadIan Campbell 🏴18h ago

@Viss the day has escaped me :( but let's find time soon, please.

i am a grumpy fucker this afternoon though, and should not expose you to that. lol

Show threadViss18h ago
@neurovagrant im down to throw shade if you wanna vent too!
Show threadInfoSecSherpa18h ago
@Viss @neurovagrant
Show threadIan Campbell 🏴18h ago

@Viss if i start venting

i may never stop

=)

Show threadViss18h ago
@neurovagrant
ᵒⁿᵉ ᵒᶠ ᵘˢ
ᵒⁿᵉ ᵒᶠ ᵘˢ
ᵒⁿᵉ ᵒᶠ ᵘˢ
ᵒⁿᵉ ᵒᶠ ᵘˢ
ᵒⁿᵉ ᵒᶠ ᵘˢ
ᵒⁿᵉ ᵒᶠ ᵘˢ
ᵒⁿᵉ ᵒᶠ ᵘˢ
ᵒⁿᵉ ᵒᶠ ᵘˢ
Show threadkajer :notverified:18h ago
@Viss @neurovagrant we venting here???!?!
Show threadKrypt3ia18h ago
@kajer @Viss @neurovagrant Just what the AI wants....
Show threadschrotthaufen20h ago
@Viss @neurovagrant Wtf is “harness engineering”?
Show threadViss20h ago
@schrotthaufen @neurovagrant so you know what prompt engineering is, right?
Show threadschrotthaufen20h ago
@Viss @neurovagrant Yes
Show threadViss20h ago

@schrotthaufen @neurovagrant so harness engineering is tuning 'the thing you use to talk to the llm' instead of 'wordsmithing your prompt'. because the harness itself does a lot of the heavy lifting.

thing of stuff like claude code, crush, opencode, openclaw, nemoclaw - these things all talk to the llm on your behalf and handle a bunch of the heavy lifting, so "your harness" can be way more effective than "your prompt"

Show threadschrotthaufen20h ago
@Viss @neurovagrant Ah that makes sense. Thank you for the explanation.
Show threadRootWyrm 🇺🇦22h ago

@neurovagrant bluntly, these people are delusional.

I have seen LLMs stacked against well established ML systems. Because the decision was made (incorrectly) that LLMs would be 'cheaper.'

They were quite literally multiples more expensive. And the results went from 97% accuracy to <70%. Getting it anywhere near the same level of accuracy would multiply costs again.

Show threadRootWyrm 🇺🇦22h ago
@neurovagrant and it wouldn't surprise me if most of them have no existing ML, or their ML was ineffective nonsense.
So they're quite literally incapable of seeing that they're wasting multiples for below acceptable results. Or they've gone all-in on the psychosis thinking LLMs are good at regexps (they are absolutely not.)
Show threadIan Campbell 🏴22h ago
@rootwyrm ***Also - please save the snark. I'm tired, and this is a genuine, if hesitant, ask.