This person gives elon musk $8 a month
When I was much younger, a verified user meant that you were someone important. :-/
The unifying thread across all high-control systems, from the earliest monarchies to industrial barons to modern tech-accelerationism and the maga movement, is the exploitation of manufactured fear and artificial scarcity to consolidate power and wealth while demanding unquestioning tribal loyalty. They also offer subscriptions for protection from the issues they made too, some called tithes, and others this way.
Also, someone has to run the AI and know how to fact check the output and know what prompts to input. And also be able to talk to customers and those people generally want to talk to a human who actually can understand them. Also, no in office workers means a real estate collapse. Everyone had to go back to the office urgently for the economy and for the real estate but also get fired cause of AI. The logic did a little hop/skip there.
I had to take mandatory AI safety training at work. One thing that they kept stressing was how raw data shouldn’t just be handed over to AI tools.
If you wanted to analyze customer orders, you should first redact the dataset by removing customer names, addresses, and any other identifiable information before giving it to the AI to ensure company secrets aren’t leaked.
Is the tool so amazing that it’s worth the risk of exposing trade secrets every time you use it? My company leadership says yes!
And the reason this is even tolerated is only because most people don’t understand how LLMs work. Like politicians and their “kill switch” idea if an AI “goes rogure” for example. That there’s like one persistent Claude entity, living in a big room at Anthropic HQ remembering and learning things from all the users it interacts with… Of course it’s not that, it’s an instance of a program that launches and runs the same function every time on whatever inputs it is fed. Sometimes it is set up to feed back old inputs each time so it acts like it has some memory of what your instance previously did. Training a model is sort of like creating a hologram and prompting it is sort of like taking a photo of the hologram from a certain angle.
They could of course train the next version of the model on everyone’s prompts to the previous version- and probably are. But that’s in no way necessary, and is already sort of a desparation move when they’re running out of input text.
If the trade secrets / PII you’re foolishly sharing with it are getting leaked that’s not an accidental, inevitable consequence of how AI works. It’s because the company hosting the AI is greedily harvesting and storing all your inputs, for no other reason than you agreed it could in the TOS you didn’t read, and then later oopsie they failed to secure it because they don’t really care and know no one will hold them accountable. Yor prompt is the same as any other text you sent in an input form to any website, the company has it and can decide to only use it for what it has to- or for everything it can get away with.
Some companies now have AI trainings, that are AI-generated, like AI-generated avatar with AI-generated voice tells people about AI.
What in the dystopian cyberpunk hell is that. 🫠
Even more fun. My kid’s school district had a presentation that using AI to do work was cheating and should be avoided. That presentation itself was GenAI slop…
I was promised neon cities and brain implants, dammit.
My last job training was blatant Ghibli AI slop through the entire thing. We get that instead
I’ve got some good news and some bad news about the brain implants…
You realize de-anonymizing data is a well researched, effective workflow at this stage? I would be entirely unsurprised at OpenAI and Anthropic using it. Meta are masters at it.
I’m constantly shocked (and not shocked) at how companies are doing this and at the same time aren’t hosting the models themselves. Running deepseek on even the cloud is much cheaper, it’s your vms and your data. Or god forbid your own infra. Seems way safer, and yeah you have to maintain it, but it’s still your data.
Of course they fired everyone who knew how to stand up all of that so it probably isn’t even on their minds.
The whole point of AI is to get everyone locked into the cloud, renting forever, without the knowledge to escape.
The frontier vendors are great hype people.
Local Deepseek (or frankly, Eliza running on an 8086) is probably perfectly fine slop for 95% of your slop needs, but you’ve gotta have the biggest and shiniest slop. You clearly need a LLM powerful enough to invoke the apocalypse to polish that marketing budget spreadsheet, or you’ll be left behind.
Right, I’m even amazed that more businesses aren’t just using older gen models for slightly less ‘genius’ at a fraction of the cost. Maybe everyone’s buying hype, or maybe there’s a hiccup in enterprise scale contracts that I’m not familiar with.
One of our executives has been going off on how the cheapest model from the vendor is plenty good enough for anything the company might want to do. Hasn’t made the leap to just running it on-premise, but certainly already trying to still say “it’s really important” while “don’t spend so much money” at the same time.
Modern business is addicted to outsourcing any aspect of business that’s not a “core competency.”
They pay Amazon or Microsoft run the data centers. They pay Accenture or Infosys for enterprise management applications, and so on.
They expected to pay their LLM vendors to build and maintain the LLMs, only to find out that they cost too much and don’t really work like they were promised.
The irony of that post having all the markers of being written by AI
“That’s not [x] that’s [y] [with a z modifier]“
That’s not [your girl], that a [clanker][hallucinating deleting prod]
That’s not a rock, that’s a rock lobster! 🦞
Guess where AI learned that? From humans. Particularly LinkedIn and Twitter. And this is a Twitter post.
They learned it from us. People already wrote that way. We used to call it corporate-speak or posting by committee.
Not all the markers. I don’t see an em-dash in there.
You didn’t hire a replacement—you gave a stranger with no soul the keys to everything you own.
Fixed it!
i like the reminder to not work or even answer contact from work if you are not paid though…
Brutal. I love it.
I love how the AI just stops working if the tokens are gone.
They can’t be coerced into working overtime. They can’t be threatened with being fired. They do what is in their contract, and that is the end of it.
Don’t like it? Go find an other AI that will do exactly the same.
Tbf that’s one good thing we can learn from LLMs. When the pay stops, your labour stops.
Your pay doesn’t stop when you’re on salary.
That’s why you take stuff from work!
It does stop when your shift ends.
Your shift doesn’t have set start and end times when you’re on salary. You might work 40 hours one week but 54 hours the next week, and maybe 34 hours another week. For the same amount of money.
In developed countries, your salary is for a fixed week of working hours. So if my shift is 6-14 monday-friday, I work for that time. Anything outside that is considered overtime and strictly regulated. People also have the right to refuse work calls after their shift ends.
That’s not how it works in the US. But you could argue we aren’t a developed country anymore since Trump came into power.
Anymore? I can’t imagine any self respecting country to expect its citizens to pay for insulin and cancer treatment. The USA is a developing country cosplaying as a developed country.
That’s exactly my point. Sorry to tell you but I can’t consider the US a developed country. It’s basically a third world country with a lot of billionaires and car-centric cities that look modern. But that’s basically it.
What? No. I have work hours in my contract, if I work over that I get overtime. What contract are you on?
Most salaried people in the US have hours on paper but are coerced into working more. I’m theoretically supposed to work 40 hours per week, but if the production line is down they beg me to come in. I can say no, but the people who say no don’t get promoted.
You assume minimum wage workers like me get a salary lol
Edit: in the uk, if your total pay divided by your total hours falls below minimum wage, then yes, your pay stopped and you still worked. From my management colleagues who were on salary, they still had a set number of hours they worked per week.
Strangely enough the deputy manager who worked twice my hours got twice my takehome pay, despite that I was just a sales assistant.
Fun fact, you also describe some outsourcing arrangements that my company got into.
They decided the outsource strategy would be cheaper. Then shocked Pikachu face when the company said to fuck off when they had new/changed requirements unless they brought way much more money. They did actually learn there was value in having a workforce that is inclined to “suck it up” versus the seemingly cheaper up-front outsourcing company, and spent years building back ability to do it in-house again.
And lately they’ve been mumbling that maybe outsourcing would be cheaper again…
You gave your corporate competitors access to all that.
The Amazon style of business, that offers a platform to sell stuff but copies anything worthwhile and releases it under an Amazon brand. Now coming for whole businesses.
It’s just weights bro please bro c’mon it’s federated bro no pii leaves your org bro, just a bunch of matrices of vectors, it’s totally different bro
Plus, with no employees, there’s no one with a paycheck.
No paychecks mean no customers.
We think billionaires spend lots, but they don’t. Jeff Bezos buying a support yacht for his megayacht every ten years doesn’t support an entire country’s GDP.
Its those with the least to spend who are the actual job creators. They spend every penny they earn (and then some) because they have to in order to survive.
Token costs and the permission problem cost the company itself, while eliminating paychecks is a negative externality. 99% of the paychecks that fund their business come from other businesses.
No paychecks mean no customers.
That’s never been a concern with the corporate executives. The macro level stuff is someone else’s problem.
We saw it in the 90’s when they stopped hiring managers from within, instead relying on bringing in outside consultants, and then directly hiring the ones they liked, so that there was no longer a path between the typical employee’s seat and their boss’s seat.
We saw it in the 2000’s when companies realized that they didn’t want to train entry level employees anymore, and would rather wait for someone else to train them, and then hire midlevels. Except too many employers did that and then the pipeline of future midlevels dried up in a lot of industries that were then ill positioned to survive the tumult of the 2008-2010 recession. Then something similar played out in the 2010s in certain other industries.
It’s always been buck passing. They don’t care.
this seems to be written with AI as well
It does but that’s just because we’ve had this kind of pithy+clichéd attention arresting tech bro speak for years.
The models gobbled it up and treated it as vaunted gospel because it got engagement.
But it didn’t start with ai and it won’t end with ai.
Because at the end of the day if you think the problem with everything is ai.
Then the problem with everything will be ai.
For you.
Enjoy.
Roflol OK I’m bad at it but it is kind of fun.
🤌
Had me in the first half…
It is, and I appreciate you pushing back on this.
Sounds like A.I. would be replacing CEOs very soon.
Worth pointing out that you can run models locally on consumer hardware that are as good as frontier models from ~3 months ago (and that gap is getting smaller and smaller) which solves ever single problem listed in this post
Except for the problem that the idiot managers didn’t do that originally. Also the problem that they don’t seem to be aware of how to do that, probably because they fired their IT team.
The thing I have learned recently attending industry events is that while what you are saying is technically feasible, it’s not as tantalizing as what the AI giants are promising. They are selling an all-in-one solution that was supposed to reduce costs and make skilled work idiot-proof.
To run a local model, they still need to have at least a few people on staff that know what the hell they’re doing to set it up and maintain it. Not exactly the silver bullet that they were expecting.
it’s not as tantalizing as what the AI giants are promising
I lost my job years ago to such con-men selling the idea of “outsource your IT to India, they’ll work for only $2 a day!!! Imagine the savings!!!”
And then their IT infrastructure went to shit. And any of us unemployed IT guys who called them out were labeled as racist for pointing out the people they fired us to outsource to were objectively unqualified.

Business owners need to wise up and realize every sales pitch they hear is probably a conman lying to get their money.
What are some local models worth exploring?
Qwen 3.8 has a 27b (consumer gpu) model and one that will run in 128gb. deepseek 4.1, glm 5.3 flash, and mimo 2.6 flash also fit in 128-256gb ram, and challenge models for 1-2 months ago.
128 GB RAM? In this economy ?
Does it link to the internet and do I have to let it access every part of my life?
Idk man in a surveillance state, AI seems like the most obvious way into everyone’s lives that’s ever been deployed by companies who are awful and shamelessly so.
I’ve no interest in trying to run an AI locally on my own equipment but I have set this up for work and it’s internet access is monitored and heavily restricted. It has complete intranet access, so that’s all the documentation it needs a d it can access read only versions of anything saved in one drive. It can visit websites but it doesn’t have the ability to upload anything, it can’t even send emails, it can write the email, but a human needs to press send. It seems to be working pretty well, I haven’t heard a peep from management since, so I’m guessing they’re happy with it.
For the case of local models, no not at all. Most local models are under a free license (or at least a non-commericial one) and it all runs locally on your hardware, however what you can run and how well depends on your hardware and some software fiddling. Non-local stuff if 100% a ploy to make everyone dependent on big tech even more than they already are, though
All this only applies to the cloudbased AI guys, like OpenAI and Anthropic. If you run your own open source models such as Kimi or GLM on your own or leased hardware - poof, problems gone.
I mean, not quite, but also yes.
Running your own model doesn’t solve every problem with LLMs, but it sure as hell does solve a ton of them, for way less money! Local models are great, gwen3.8 on my 3090 at home is slower, but often better than the pay to play Claude from work.
It maybe doesn’t solve ALL problems, but it solves the ones in this post: high token prices, running out of tokens shutting down the workflow, and handing over your sensitive data to corporates doing god knows what with it.
One of the better aspects of it is that it decentralizes power and cooling. When everyone has to juice up their own GPU and deal with the noise it makes in their office, you remove a lot of the problems related to the data center chewing up enormous amounts of electricity and water (or rather, if lots of people did that, it would target that problem). And that’s a lot more tolerable than giant data centers increasing the local temperature and chugging down the municipal water supply. Problem being that GPUs are now way too expensive. This will be a more approachable model when the bubble bursts (I hope).
Exactly, that’s why I gave it the “No, but actually yes” preface.
That’s what I run on my 7900 XTX and it’s about as good as Sonnet 5 that I use regularly at work. Zero reason to give Anthropic or OpenAI my money or data.
No it isn’t. Those models are trained on the big models. They will not get better if the big providers go under. Sure, they’ll keep running, but half of the reason people are buying into all this AI nonsense is that it’s going to be much much better than it already is in the future.
That’s just not true. They might steal from the American models, but they don’t depend on them.
They literally use them as base models for compression.
How do they do that exactly?
It’s a process called distillation. One way of doing it is creating thousands of accounts to prompt a big flagship model, then use the output to pre-train their own model.












