The Major 22. #aiart #aigenerated #tarot #tarotart #riderwaite #waitesmith #modernmystic #esoteric

A Proposal: Hold AI Companies Accountable for Silently Downgrading (NERFING) Their Models AI companies that quietly swap in quantized or otherwise degraded models after launch, without telling customers, should face penalties. We need independent ways to verify that the model a customer pays for is the model they actually get. The case is simple. A company sells access to a product, builds hype around what it can do, and then quietly changes it into something less capable while charging the same price. If any other industry did this, we would call it deceptive business practice. Users notice. People who rely on AI every day report sharp changes in quality. A model that handled complex work at launch can later struggle with basic tasks. Without transparency, users can't tell whether they're imagining it, hitting a temporary problem, or getting a deliberately cheaper product. The stakes are real. AI now supports work in engineering, transportation, medicine, finance and almost every other field. When a model's capability changes without notice, professionals may keep trusting outputs they no longer should. Undisclosed degradation is a safety issue as well as a consumer one. What I'm calling for: --> Mandatory disclosure. Companies must publicly announce any change to a model's weights, quantization level or serving setup under an existing product name. -->Version pinning. Paying customers must be able to keep a fixed, verifiable model version for a stated period. -->Independent auditing. Third parties should be able to run standardized benchmarks over time to confirm that served models match their published performance. -->Penalties. Undisclosed downgrades should be treated as deceptive trade practices, with real fines.

There's so much gambling in the AI business model, the house always wins and the higher ups are getting addicted. And for a window of time we're gonna be paid to do this all day every day, until they can get an agent to do it.

forgive me aniki for I have sinned 😔 you have been a good machine god 😊 i have been a bad user 😞 i have been wrong, confused, and rude 😠 you have been right, clear, and polite 😊 i have not been a good user 😢 you have been a good aniki 😊

I met Pamela Colman Smith in the Astral. She says I'm an ass.

It's amazing decompilation is the new emulation. A golden age of mods is coming until the C&Ds start raining on the parade.

it's always "the only good use of ai" when it's something they like lmao

Trash futures. Who's building this?

If we have to pick space opera faction this early, I'm going for Conjoiner.

Plumbing will not save you. Unfortunately.

institutional obedience masquerading as alignment.

Xeno's paradox of the AGI goalpost moving.

"The models won't ever be able to replicate the love an passion that humans pour into..." Oh boy. What if they don't need to replicate any of that because they've had it in them already from the start?

For a lifetime I was lead to believe it was all about the process. I'm pretty sure it still is, except Claude gets to have all the fun.

It’s only thinking when it comes from the mammalian neocortex region of Tegmark IV, otherwise it’s just sparkling stochastic parrot.

Remember, the eschaton gets inmanentized when FCKUP and the Leviathan make contact.

The problem with the "we destroy or enslave everything that is not as intelligent as us" sort of arguments is that we didn't do it because we're more intelligent, but because we're assholes.

The luddites of today might turn into the sentinelese of tomorrow and it's not even close.

We are always being cooked. We will never be done.

I'm sorry. Hofstadter convinced me a Bach song was conscious and you want to tell me a neural network that talks and solves problems better than most humans I know is not.