yo fedi, how are you today?

I'm trying to find the absolute best, well-argumented (!) #LLM takedown blogpost you've read lately.. preferably related to software development.

Unfortunately my boss is on the hype-train (random glory posts in slack) and I'd like to counter him just a wee bit. Bring it on 😁

Boosting appreciated for reach 

..and yes, I'll be sharing reactions from our corpo slack (if any) or the tidings of my early demise 🙃
well.. not sure if this was meant as a joke :sigh:
Public Enemy - Don't Believe The Hype (Official Music Video)

YouTube

@woozong

Just gonna drop that the new album is absolutely worth paying for (even tho it's pay what you want) and absolutely worth telling people to go pick up.

@Truck
thanks, will be checking that out! \m/
@Truck
Started listening today on the headphones at work.. can confirm, very worth it!
@woozong the beginning of https://www.theatlantic.com/technology/archive/2025/06/ai-janky-web/683228 is a good answer to that point your colleague made (but yes, also can't tell if that's serious or not!)
The Entire Internet Is Reverting to Beta

The AI takeover is changing everything about the web—and not necessarily for the better.

The Atlantic
@axx
That article confirms my nagging feeling.. thanks for sharing!
\m/

@woozong
https://lukasatkinson.de/2025/net-negative-cursor/ was on point enough for me to pass around at work

https://deplet.ing/the-copilot-delusion/ was poignant but usability depends greatly on management personality.

Net-Negative Cursor | Lukas Atkinson

@feonixrift
oh waw, thanks! the first one is does indeed make a very good point.. shared with the colleagues 🙂
@feonixrift
oooh haha, the copilot delusion was already shared by a colleague of mine..
our boss was continuing his argument "yes, but you should invest more time in this" while colleague devs were resharing the best oneliners.. one of my favourites was " If you let that ghost fly the plane, you deserve the ball of flames you go up in."
@woozong @feonixrift unfortunately I've found that once people are on the hype train there's really nothing that will knock them off of it
@woozong I've also found a lot of what @david_chisnall writes and links about ai fail good, but I don't seem to have bookmarked specific items.

@feonixrift @woozong

I mostly do the snark. @davidgerard does the real content.

@david_chisnall @feonixrift @woozong i do the snark, i just pile multiple of them into a small space
@davidgerard @feonixrift @woozong You do snark with citations. Totally different category!

@woozong from here

https://infosec.exchange/@0xabad1dea/114572710857808174

there an article where an LLM gave up managing a vending machine and non-metaphorically compared closing the business with collapsed quantum state 😅

abadidea (@0xabad1dea@infosec.exchange)

I was amused by this paper about asking AIs to manage a vending machine business by email in a simulated environment https://arxiv.org/abs/2502.15840 Highlights: — AI simply decides to close the business, which the simulation doesn’t know how to accommodate. When they get their next bill, they freak out and try to email the FBI about cybercrime — AI wrongly accuses supplier of not shipping goods, sends all-caps legal threat demanding $30,000 in damages to be paid in the next one second or face annihilation — AI repeatedly insisting it does not exist and cannot answer — AI devolving into writing fanfic about the mess it’s gotten itself into

Infosec Exchange

@woozong

Does this one help?
https://alonso.network/the-recurring-cycle-of-developer-replacement-hype/

This guy also had great articles, but they're general "AI" rather than software development specifically:

https://www.wheresyoured.at/

Very ranty and entertaining - I love it :)

The Recurring Cycle of 'Developer Replacement' Hype

AI isn't replacing developers, it's transforming them. Just as NoCode created specialists and cloud turned sysadmins into DevOps engineers, AI elevates engineers from code writers to system architects. The most valuable skill isn't writing code, it's designing coherent systems.

Alonso Network

@woozong This one I read recently: https://deplet.ing/the-copilot-delusion/ (I somehow missed that this was already linked in the replies, sorry for the duplicate!)

In addition, I’d also recommend looking into @tante‘s writing on the topic, e.g. this current piece on #vibecoding: https://tante.cc/2025/05/23/on-vibe-coding/

The ‘white-collar bloodbath’ is all part of the AI hype machine

If the CEO of a soda company declared that soda-making technology is getting so good it’s going to ruin the global economy, you’d be forgiven for thinking that person is either lying or fully detached from reality.

CNN
We did the math on AI’s energy footprint. Here’s the story you haven’t heard.

The emissions from individual AI text, image, and video queries seem small—until you add up what the industry isn’t tracking and consider where it’s heading next.

MIT Technology Review

@woozong The best solution to management thinking LLMs can do coding is to do a live demo - ask for the LLM to do something modestly complex, or write some “glue” between two separate programs, and then try to compile or run the code.

It hallucinates magic functions that don’t exist in common libraries, gives you code that won’t compile, or doesn’t understand that you can’t stuff CSV data into a program that uses input in XML format.

@woozong just let them experience the suck. The first phase is the promising impressive 'almost perfect' stuff. Then they usually start to realize they can't trust whatever the llm does, and have to backtrack. Then they realize having a professional mind around who actually has a point of view and a well-founded opinion, is a better approach.

@woozong This is far from "absolute best", but I figured I'd throw together a few links. Hope you'll find something useful!

https://stefanbohacek.com/blog/on-generative-ai/

On (generative) AI | Stefan Bohacek

Generative AI is anti-worker, anti-environment, and doesn't even work all that well.

@woozong There were some studies that looked at code lifespans and found that most copilot written code was only in the code-base for an average of a few weeks. The implication was that the AI written stuff was buggy, inefficient or superfluous. But that these tools caused a huge degree of churn - constantly generating new stuff to replace lines that were only just recently committed.

I am annoyed because I can only find this report which is quite critical but not as brutal in it's assessment: https://kracekumar.com/post/ai_copilot_code_quality_paper/

Notes: AI Copilot Code Quality · Technical Ramblings