Post

Half my job has turned into AI related work, due to being a data wrangler. There are a lot of legitimate uses. But I'm also seeing a LOT of ceding work that isn't best served by being done by an LLM.

This, but folks seems to be very binary in their approach: either LLMs run my life or LLMs are the antichrist.

Slippery slope from using an LLM to sort/cat data to slopping yourself. As with everything else free on the internet YOU are the product and addiction to it is the point. Agree to disagree with you and OP. Stay away from this shit unless all it does is check your PowerShell for errors.

That tiktoker who came up with "slopper" to describe someone who constantly talks about using LLMs was truly inspired That word/framework is pure gold & I'll be using it daily

LLMs suck ass. They’re cloying and annoying and wrong a lot.

using tax lawyers?? huh?

Partners company have a hiring freeze while they look into how they can utilise LLMs. Hoping it comes back to bite them, has pissed off many who have been denied leave & transition to retirement. Real possibility some will just quit which is a problem as takes 6 months+ to train replacement.

Yup, use of AI is now a thing to detail at my company. I had to invent a use for it and I’m not actually going to do it

Very much yes to this.

Meanwhile if I were a hiring manager, an employee distrusting LLMs would be a MASSIVE, HUGE green flag because that means they understand what LLMs are and why it's a terrible idea to lean hard on them for your work.

I’ve literally only found them useful in drafting resumes and cover letters because ironically the less human you sound, the more likely you are to get through to a person

It rules that some of the most credulous Business Idiots in the world decided to try to reorient the entire tech industry around a cross between a mildly useful word plugin and a magic eight ball, love this for all of us

It's like some sort of highly specialized mining equipment, something most people don't need and wouldn't know how to use. It's being sold as a replacement for screwdrivers, a task it absolutely cannot do. But for the most part, I agree with you, it's largely useless and not worth it

Converting random data to json.

Evil regexes… fuck, I remembered

Please do not. They will end up doing crazy unexpected matches. That is a place *not* to use LLM. Crappy FastAPI interfacing to internal libraries, sure. Or generation of fast & cheap dashboard sites.

I make my own regexes and never thought to have an LLM do that. I use them to extract tabular data from PDFs, except often they are confidential

Seriously, like mostly transforming structured data not much else in my domain

.*

nvr seen sum1 pluralize regex b4

10/10 would use for regexes, they annoy me

lmao exactly

LLM can organize these 100 messages into similar buckets so I don’t have to do it myself. I still have to double check all of it and make sure it makes sense but it’s easier than starting from scratch

I would pay actual money to hear any one of these AI hyping fops to give an honest breakdown of costs and benefits. Like here's the real cost of compute in 2025 dollars, and this is the real use case that would at $15 minimum wage cost X. But it's all just incomprehensible gobledigook.

Don't get me wrong, I tremendously enjoy trying to get AI image generation to add cookie monster or oscar the grouch into scientific imagery, but it's not exactly worth the real cost (be it old growth forests or poisoned minority neighborhoods) to justify keeping it.

I refuse voluntary cognitive decline (bullshit machine edition).

I still personally refuse to use them for anything out of protest. I know I am robbing myself of some use cases but I still personally believe the energy costs and pollution just isn't worth anything I encounter in my day to day

And you never learn how to do that if you let the AI do all your thinking for you.

On the flip side investing time to learn how LLM actually works is a good brain exercise and helps sharpen cognitive skill.

Anyone who got a Ti-89 way too young and then failed to learn much of Calculus (while of course still passing due to said Ti-89) will understand this.

Agreed. I really see it as so spotty and more of a pain sometimes to use for anything complex. Basically use it like for assistant low level internship work and I think you're fine. I like using it for making simple Excel charts or just drafting an email before I go back and revise that shit again.

Unless you have a human knowledgable on the subject vetting a machine's output, you can't trust anything that the machine confidently says is right, or even real. I can hallucinate perfectly well by myself, thank you; at least I can tell when I am hallucinating.

I would call it a "bullshit machine", in the sense of bullshit described in Harry G. Frankfurt's treatise "On Bullshit". The LLM can neither lie nor tell the truth because it doesn't have intention. It's bullshit in its purest form.

(I probably botched it, a lot's been happening in my life since I heard this talk and frankly keeping my eyes open and putting one foot in front of the other has been tough work. But I still do insist on knowing things. Also, Ken is a great, engaging speaker, as it turns out.)

Shameless self plug but here's my Substack for those interested in em-dashes (and tells of a NatSec DOGE'D worker) open.substack.com/pub/daggeran...

what the actual fuck is about all I can say

I almost feel even calling it the "Lying machine" gives it more credit than it's due, since lying at least implies intentionality and understanding.

We still have not come to any terms with AI with how to deal with its implicit bias, or anything that seems to be hate speech, crime, or violence. Where’s the ethics in these stupid things??

I remember someone once posting "I want AI that will do my dishes and my laundry to free me up for art and music, not the other way around."

and by interesting I mean that sucks

But, but... we MUST work 996 workweeks so that our AI overlords are free to write shitty, derivative Bulwer-Lytton novels, and generate Thomas Kincade-style paintings of women with five boobs.

So, damned if you do, damned if you dont?