evilstevie @evilstevie@mastod1.ddns.net

I like this post from 2017...

Edit: here's the related news story
https://www.bbc.co.uk/news/world-europe-30100756

See you tomorrow morning in the Corgi Cafe in San Francisco 9 to 11 for a "Help me fix this" meetup.
https://luma.com/weared-zzo9

Jeebus this entire piece is a phenomenal piece of writing.

Just a reminder that AI detectors don't work. You don't fix one broken machine with another.

I am shocked. Nobody could have seen it coming. Nobody, I tell you!
https://futurism.com/future-society/college-critical-thinking-ai

> As one New York financier told Financial Times journalist Gillian Tett, new hires who were seen as “AI natives” are turning out to have alarmingly shallow ideas. So much so, the anonymous finance worker admitted, that his firm now actively avoids seeking out AI-literate STEM graduates, and opts to comb through humanities students instead.

Oh, the humanities!

#AI

A big issue with LLMs:

It makes some people believe they are more competent than they are.

Example: this company sent me some spam. At first I thought it was a scam. Then I dug into the company.

And I realised: oh. No. This company is not a scam per se. No, these people actually believe their own bullshit.

The CEO ran a web server as root. It got owned. He posted about it on LinkedIn. He used Claude to "fix" it. It was not fixed.

Now you see, before LLMs, if you didn't understand a domain, you'd hit a wall very fast. Documentation would use terms and concepts you didn't understand. Tools would require interpretation. You'd have to ask somebody more experienced. That friction produces a very useful insight: "Oh. I don't know this domain very well." So you'd find someone who did, or learn.

The thing about LLMs is that they can remove an enormous amount of that friction. That friction is not noise, it is signal. That friction is telling you that you need to learn the domain you are working with. Removal of this friction is a very, very bad thing.

The LLM can hold your hand, do a lot of heavy lifting for you, use a lot of tools, run a lot of commands, and make things appear to work.

And the thing is, all of the individual outputs along the way might be pretty good and it might look correct.

The issue is in aggregate. Before you know it, you've built a house out of sand.

And worse yet, people can emerge from the experience with a dangerously incorrect higher-order conclusion: "I am now a domain expert in this."

They are not a domain expert. They understand absolutely nothing about it. The LLM did everything for them and it's wrong. And worse yet, they haven't reviewed it, because they lack the competency to review it. Therefore, they assume it is fine. But it's all incorrect. A true expert would see that it is wrong, and correct it, or push back on the LLM, or do it themselves.

And this is how you wind up running web servers as root and getting owned.

I keep finding myself in situations where I explain my job to someone new to infosec and they look a bit disappointed that I emphasize active listening skills rather than increasingly difficult topics in mathematics or programming languages to study.

I sometimes joke that 50% of my job is "a weird form of group therapy", because I don't know how else to describe it to folks who never developed the requisite social skills.

Technical competence matters, but is not alone sufficient to carry you in most workplaces. Being an irritating asshole all the time is a net negative. Being an irritating asshole when you need to be is still useful, of course.

But I'm not exactly an extrovert, myself, either. And there are plenty of everyday social norms that frustrate the hell out of me.

For example: I cannot relate to people that routinely buy celebrity gossip magazines. No amount of communication skills will bridge this gulf. Y'all are a lost cause to me.

Well ouch.

if a hive finds itself in uncharted waters, is that undefined beehaviour

"influencers getting bitten while taking shark selfies" seems like a problem that will sort itself out

really funny how slop fiends are like "yeah it's slop but have you considered i don't care" and we're like, yeah, we noticed that you're a careless fuck, that isn't a shocking revelation

It must be a very worrying time for anybody who has dumped a body in a reservoir.

I don’t know how else to say that if you are against using any LLM, then you should stop using AI checkers of any sort. Not only are they just another tool to slurp up text, but every bit of writing with common words and phrases will be flagged as generated when it’s not because every LLM has common words and phrases in its training data. Lastly, every one of these checkers is designed to eventually get you to pay the tech company money so they can pretend to make your writing more human. They do not work the way you assume.

I know this is radical but we may just have to trust artists! I know, what a shock.

Besides, these AI losers can't shut the fuck up about using their generator so it really isn't hard to find people that use these things if you really wanna point the finger at someone.

#AI #LLM #NoAI #AISlop