Emad Mostaque and UC Berkeley's Jason Lee debate if the US can block open-source models matching GPT-5.6 performance

@kentcdodds This, in my opinion, has been Dario’s aim the whole time.

@jpschroeder @kentcdodds An entire book was written about what Anthropic is currently doing, George Stigler’s The Theory of Economic Regulation.

Will above fable level open source models be made illegal in the USA

@kentcdodds They choke the hardware. We wont be allowed to own computers any more watch.

Hyperbolic sure but this is the only play they have

@kentcdodds They'll just take sites like HuggingFace offline and commandeer any repos serving advanced LLMs. It's a lot easier for them to cut out the head off instead of trimming each hair individually.

@EMostaque Hard no! That would be like the USG saying anyone other than US citizens is allowed to use Fable-level AI! It would mean total domination by China and the fall of America.

@kentcdodds The good news is, they likely can’t. Not practically. They can bitch and cry and scream illegal and banned. But you can’t realistically hold back peer to peer sharing over the internet of widely distributed models. That’s our only hope for freedom of access to ai as well

@kentcdodds They're hardening critical infrastructure before that happens. How hard is that to understand?

When open source Mythos 5 models come out, they'll have to contend with defenses built by Mythos 6, months prior.

@kentcdodds They just need to tell the neo clouds and hyperscalers to restrict them, and take huggingface etc offline. There will still be some illegal usage but it would be effectively dead for real businesses.

@kentcdodds Chinese models are so bad they will never catch up, but again SOTA models usage is getting so ridiculous you will burn a 20$ monthly plan in a few hours, they are not even usable anymore, So we might be stuck using trash Chinese models.

@kentcdodds

When they say "turn it off or else" but you're open source

@kentcdodds Answer: not really but they will try They will outlaw Chinese models first Eventually they will take down Huggingface Then we will switch to torrent and they will go after that Then they will increase criminal penalties for simply possessing outlawed models All counter-productive

@kentcdodds Why do you think they are enforcing OS level age verification soon. There’s no escape. Make the silicon or be controlled it seems.

@kentcdodds They cannot without censoring the internet itself in which case we become china. Everything we say we’re not. And the Republican Party can never again say they’re the party of individual freedom and autonomy.

@kentcdodds Yeah, this only buys them a few months I think

@kentcdodds This haphazard whatever some rando bureaucrat in the Trump admin decides method is only going to hasten the open weight and foreign based models.

@EMostaque The US basically can't ban Chinese models. If the rest of the world has only Chinese AI the US loses. The US loses even if Chinese models aren't banned and they're better or nearly as good and much cheaper than US AI. Holding back frontier models is putting a lot at risk.

@kentcdodds Restrict the hardware they need to run on and forbid the US cloud companies from hosting them?

@EMostaque There was a time they fear mongered that electricity. Especially, AC, was to dangerous for domestic purpose. But in reality the problem was solved with better technologies and standards of operation. So the USG ban is basically them being lazy on tackling issue properly.

@kentcdodds Thats the only cause why DRAM prices increased drastically. It's all made up stuff to stop (uncontrollable, uncensored) local AI being too potent. Just imagine having a consumer GPU with 128 - 512GB VRAM.