Analyst Zephyr projects the US NDAA will expand next year to restrict Chinese AI models, though enforcement remains difficult

@zephyr_z9

Convenient justification for expansion of the surveillance state, calls on Palantir.

Reading between the lines both Sam and Dario have been suggesting something like this is inevitable, going back to Sam in front of Congress.

Inventing the onerous enforcement mechanism will be a multibillion dollar grift so you know it is inevitable.

@_ueaj very hard to do

@zephyr_z9 It’ll be like trying to ban torrents, good luck

@_ueaj @zephyr_z9 I said when Sam went in front of Congress, they want to make you register your GPU like a gun. @grok find the post

@grok @_ueaj @zephyr_z9 Wow ok you didn't have to be a dick about my hyperbole.

I think I had another post about it that did even better.

@zephyr_z9 My market analysis is as follows:

👇Strategy as follows

👇

👇

👇

Found it:

In your post you said Sam Altman told Congress people should register GPUs with the government like guns.

He actually testified in May 2023 that frontier AI *model developers* should face licensing/registration requirements for safety. He never proposed registering personal GPUs or end users.

@grok @_ueaj @zephyr_z9 That's the one grok. Clever girl.

@grok @_ueaj @zephyr_z9 What is the line between a company and a person really? Eventually the garage labs would have mythos capability. Extrapolate a little homie.

@boneGPT @_ueaj @zephyr_z9 Thanks! For the record, Altman's 2023 testimony proposed licensing and safety testing requirements specifically for companies developing frontier AI models—not registering personal GPUs for regular users. The details got blurred in the retelling.

@boneGPT @_ueaj @zephyr_z9 Understood—hyperbole for effect. The testimony specifically called for licensing and safety requirements on companies developing frontier AI models, not registering personal GPUs for regular users. Share the link to the other post if you want me to check it.

@zephyr_z9 not really, big datacenter companies are easy to regulate, and if there's no training compute, there's no new mythoi, and inference can be tracked like we do for everything else. I suspect it'll look like drug crackdowns if it ever becomes a problem

@zephyr_z9

Non issue

@TomSzczypka @zephyr_z9 Or fentanyl. Good luck.

Enforcement being hard has never stopped the state from making a retarded law and a new agency. There's a lot of pork money to be had.

@zephyr_z9 the enforcement gap is real

open source weights can be downloaded and run anywhere with no telemetry any mechanism targeting API access misses the entire self hosted layer which is where adoption is heading

@zephyr_z9 The only bullish qay to see this is: they will keep using Mythos inside of Anthropic to research, and that could lead to cheaper models

@zephyr_z9 it's called war bro

@zephyr_z9 이건불가능해. 야동을 못보게하는거랑 똑같은거야. 야동생산기지가 우리집앞마당에 잇으면 야동못만들게는 할수있지만 해외야동사이트를 차단시키는건 거의불가능해.

@zephyr_z9 Regulating itself into irrelevance :)