Minh Nhat Nguyen says government bans on advanced AI models are driving demand for decentralized, open-source alternatives

@citrini welcome to the club

this ban has prolly reduced near-term hyperscaler demand, but significantly increased demand long-term from diversified sources. you can't just give ppl a week's taste of next-gen models, pull it and expect others not to try harder as a result.

@citrini @zephyr_z9 What’s $NVDA worth if even 25% of market goes local hardware?

@citrini How does this matter in practice? If they ban a closed source model they will ban comparable open source models. No one is risking federal prison for a few extra llm iq points.

it's the same reason i think a compute ban won't work in the long run. actively incentivising millions of engineers to leapfrog, will put things much further out of your control in the long run.

@Leaf0Q1 @citrini You cannot practically ban open source models running on silicon on your own machine

@citrini considering a local model will cost you 1/1000th of opus for the exact same specs, i’d say claude was never the reasonable option

@citrini self host a hermes agent while ur at it

@citrini interesting concept question of ability to implement? surely this opens up software for manipulation

@citrini a0t

@migbassi @citrini everything is computer

@imthewalruz @Svaghost @Leaf0Q1 @citrini you literally can't enforce it

@citrini Data centers are so much more efficient and fast, it will never be economically viable or even possible to achieve same efficiency/performance with frontier models on local hardware Today's Fable will be tomorrows Haiku "Bottleneck" is still models are too stupid without guidance

@citrini Detailed strategy🚀🍀 🔻

@citrini Detailed strategy🚀🍀 🔻

@citrini yeah but we all need our own data centers that way

@citrini Just buy semis bro

@citrini Ticker please thanks

@algoflows @citrini Frontran him 😂

@Svaghost @Leaf0Q1 @citrini I don't think that's true. Just depends on how willing you are to enforce it and if the penalties are high enough.