Sure, your vibe coded app is great. So I told Claude to look at yours and make one just like it but now I control the code of that one. Did I plagiarize your plagiarized work? Will I have mine plagiarized?

Boo hoo. Fuck you. No honor amongst thieves.

#AITeaParty #noai

I've had a big few days. I've made changes to my blog to support the human.json format and added an AI statement to my blog. I've also today farted around with tags.pub, registering this account as one from which tags are pulled and following a few tags myself.

Earlier today, I happened to run across the AI statement from Beto Dealmeida, the guy behind the human.json project. While he is all about asserting the content of his blog was 100% human authored, he built his blog with Claude.

https://robida.net/ai

I guess this is like finding out that you are more hardcore and straightedge than Ian MacKaye.

#noai #AITeaParty

Over the weekend I ran across the Human.json project. This is a protocol so lightweight it practically doesn't exist. It's a JSON file you can put on your website in which you can vouch for other sites as being completely created by humans with no LLM involvement. This allows other tools in which you explicitly trust a site to show you sites that have a transitive chain of trust from the one you picked.

I had one of those "Oh cool" manic episodes. I learned about it and within the hour I had it up live on this site. I have a pretty small pool of vouches right now. I'm not planning to expand willy nilly. I started to add other sites but then I realized while 99.9% sure they have no LLM in their creation, that number was not 100%. I will only add someone to my vouch if they directly tell me convincingly or make a clear public statement renouncing LLMs in the creation of the site.

It's a small step, but a step. After yelling at the incoming tide, it's nice to have something concrete one can do. If you want to vouch for me, I promise that Evil Genius Chronicles has no generative AI anywhere in the chain. As I am finding out various tools are using LLMs in their code, I am removing them. I will keep my hands as clean as I can in this filthy business.

Good luck out there.

Imagine if some techbros told you they invented a robot. Instead of you going to the gym, it can go for you and do all your exercises with perfect form. You would think they had spectacularly missed the point of what automation is and is not supposed to do.

This is exactly how I feel about 99% of the LLM use cases. We don't need a robot doing the things that skills us up, gives us the wisdom to improve things. It's dumb, and obvious that the goal is to make us helpless. Train the LLM on our expertise, make us forget it then sell it back to us at a hefty markup. Fuck all y'all.

#NoAI #AITeaParty

The more I think about it, the better that whole "Replace the word 'AI' or 'LLM' with 'cocaine' " idea is. The analogy is so good.

You use it because everyone else is and it makes them more productive in the short term. You get a burst of stuff done. It gets expensive, then so expensive it is a problem. In the short term lots happens but in the long term, everything gets worse and worse. Quality of work suffers, you as a person and your skills suffer. Eventually, the situation is too bad to keep the plates spinning and everything crashes.

The only problem is that with an individual on cocaine, that crash can be followed by rehab and straightening out. My guess is that there has already been so much damage done in skills atrophy, brain drain, bullshit injected into codebases that it will take years or decades to unwind. The longer the frenzy, the worse the damage.

This is exactly the same sort of existential crisis as climate change, playing out the same way, perpetrated by the same kind of people and running concurrently. Welcome to your future. We are all fucked.

#AI #AITeaParty #CocaineIsAHellOfADrug

I attended a local developer event, my first since retiring. This is exactly the kind of group we were trying to create 20 years ago when my peers and I were organizing conferences and meetups. It's in a great space by great people, much better than a non-tech hub beach tourist town deserves. This is not about them, it's about the very depressing technology zeitgeist. I won't even link the group because I don't want to cast aspersions about the fine work they are doing. I cannot stress too highly they are doing nothing wrong.

It was a lunch tech talk about APIs. The first 30 minutes were quite interesting to me and then the talk drifted to Claude and the various LLMs people use to deliver work. Once there, it never strayed. I grew increasingly checked out but tried to politely pay attention and figure out if any of this had value to me.

At the end of the event, I asked for a specific recommendation for a kind of open source software. To be honest, asking this one question of fellow human developers was 80% of the reasons why I drove the 30 minutes to downtown. The only answers I got were "Have Claude code it for you. " To me that is worse than "I don't know" which is perfectly acceptable. To an LLM enthusiasist, it is the answer to literally any development question so I find it kind of pointless to say out loud.

It's not them. I expect every single software engineer between 22 and 50 is in exactly the same spot. I just found that it really drained the life out of me. Not one of these enthusiastic developers thought they were building a future in which they were irrelevant. They were happy they get more code created per day. I wrote down the phrase that jumped out at me - "for reasons I don't fully understand". This was in regards to why token usage is lower doing things via CLI than API but it can be generalized.

Whatever decisions were made to create the code, starting at infinite possibilities down to the final source files, no one responsible for it understands why they were made. They are happy to have something that wiggles when you poke it. It passes the test cases that Claude also wrote, so great. I asked directly about this and was told it was fine, so who am I to think this sounds like the beginning of a long slow catastrophe?

To paraphrase an analogy I posted on Mastodon: Imagine your boss asks you to dig a hole. You start with a shovel but then realize that you can do the job much faster with a backhoe. You dig the hole in 10 minutes instead of all day. Your boss pats you on the back and says "Great work! Now get in." and starts filling in the hole. That is what coding with LLMs is to me, using your well-meaning efforts to build a future that doesn't need you.

If I wasn't already retired, I would have just quit today. It makes me want to never touch a line of code again. I guess by wanting to design, write, test and understand my own code I'm the tech equivalent of a guy in a leather apron doing wood work with hand tools. Maybe it's for the best I get out of the way, while the human centipede of code continually gets redigested into forms that are higher percentages of shit. Let younger people make the best of this world while I wait to join the COBOL and Fortran programmers on the scrap heap of history. Me along with all the other dinosaurs and their IDEs that actually touched keyboards and went to school to learn about the various algorithmic complexities and other arcane arts that no longer matter.

Goodbye cruel tech world. I wish I could miss you more.

This is largely the path I am also taking. I buy practically nothing from Amazon anymore. I deleted my Facebook account. I am de-Googling to the extent possible.

All of these companies are actively hurting modern life. I am trying to extract myself from the shitshow. It is exactly zero fun and it makes every little thing a huge pain in the ass. Still, I find it worth it. Fuck the Magnificent 7 and the less Magnificent 50 and all the AI bubble companies and all the fascism enablers who are willing to destroy democracy for a 3% reduction in their corporate taxes. I wish them all into the cornfield.

https://ryanfb.xyz/etc/2026/02/23/i'm_getting_the_hell_out.html

#AITeaParty #FuckFacebook #FuckGoogle #FuckFascism #FuckAmazon

Here is a thing that only recently set off my alarm bells. I see people talking up their vibe-coded project like this:

"In only a few minutes, I had a full project filled out and completely tested."

When they talk about the time scales, it makes it obvious that the test cases had to have also been generated by the #LLM because a human couldn't write the tests for a non-trivial project that fast. So, these people are content that they have a proper project because the code generated by the LLM is passing test cases generated by the LLM? Holy fucking shit!

I could **almost** buy machine assistance if you sat down and defined your functionality, wrote up the test cases that would constitute proper behavior and then let the LLM code until it passed **your test cases**. Defining test cases is hard. Making them complete requires care, attention and some deep thought.

Literally anyone can dash off bullshit ones that test the positive case and a trivial error or two. Really exploring the possibilities of what can go wrong is serious work. Because different languages handle different things (think about booleans in strongly typed languages vs the "truthiness" of Javascript for example), this is not a one-size fits all process. Training sets that worked perfectly for one project might shit the bed for another.

When I see people talking up the productivity, I'm now paying special attention to this part. More important than reviewing the emitted functional code is reviewing the emitted test code. That is where the seeds of catastrophe lie.

#AITeaParty

RE: https://esq.social/@annmlipton/116108242514723130

Take a machine that generates the probable next word based on the previous. Now train that on the collective output of an unjust society. It will provide unjust recommendations.

File this one under "No fucking shit, Sherlock!"

#AiTeaParty #LLM #FuckAI