Navigation überspringen
This explains a lot about why Anthropic hates Chinese models and open source.
0:00 / 0:56

This explains a lot about why Anthropic hates Chinese models and open source.

Abu

@abuchanlife1410 Follower

Auf X besuchen

34.309 Aufrufe

This explains a lot about why Anthropic hates Chinese models and open source. Anthropic’s Nick Marwell says frontier AI could eventually require $10B, $100B, even $1 trillion training runs. His argument: if a company spends that much pushing the frontier, competitors shouldn’t be able to distill the model and ship a cheap copy weeks later. That’s the real fight.The problem is that not every Chinese model is just distillation. Chinese labs are spending serious money on training, building their own techniques, and getting extremely good at optimization and efficiency. And as models keep improving, I think capabilities will increasingly converge. The gap between the top labs may become much smaller, which makes protecting one company’s frontier advantage much harder.

16 Kommentare

Kommentare aus dem Originalbeitrag auf X (Twitter)

  • Profilbild von Yann LeCun

    @ylecun

    - training frontier models is expensive - distilling a frontier model is cheap - this market force alone tells us that frontier foundation models will be free/open. - there are others.

    184 Likes

  • Profilbild von Flavio Parenti

    @FlavioParenti

    Oh... so now you want to protect IP. After having it fully scraped for free...

    3 Likes

  • Profilbild von Skylar_artemite

    @RuidoBarcelona

    >>if a company spends that much pushing the frontier, competitors shouldn’t be able to The market players (including the public) doesn't care what some company thinks they should do

    1 Like

  • Profilbild von Crypto Yoda

    @cryptoyoda42

    You can't just raise billions, corner the GPU market, hire a ton of AI researchers, and then declare open models are dangerous bc you can't have your expected returns. It's simply a forced argument to say they miscalculated how fast competition would take to catch up.

  • Profilbild von Ab-E-Kawser(Arnab)

    @KawserE16156

    Frontier companies can delay their model release and can distill their frontier model to produce cheaper model

  • Profilbild von Hari krishna Sharma

    @iamharikrishna1

    what if they chose not to release models could we do anything at that point?

  • Profilbild von Andre

    @_Andre_Ellis

    This is true from a business perspective… but the argument fails when you ask where the Frontier companies got their data… And now that there will be an explosion of startups in the AI space, the rules suddenly protect those companies? We can’t use everything available to us?

  • Profilbild von Sof t.

    @sophsbeats911

    Great point!

  • Profilbild von RayRay

    @lc29b50

    Fun fact: You can’t distill your way to a 3-trillion-parameter model. Let’s be honest: this is about monopolizing the market, nothing else.

  • Profilbild von RV

    @renoirvieira

    fuck anthropic

  • Profilbild von Tom Maiaroto

    @tmaiaroto

    ...and a lot of money to run the inference as well. I very much disagree with the make the model bigger and bigger strategy. "Infinity" isn't a strategy.

  • Profilbild von Assaf Petronio

    @AssafPetronio

    compete or get eaten, dont play dirty

  • Profilbild von andy

    @1a1n1d1y

    putting a multibillion dollar water bottling plant on precious/sacred lands is unfair because people can just filter tap water?!

  • Profilbild von VisiveAI

    @VisiveAI

    Train is expensive, distill is cheap — that market force alone is why LeCun thinks frontier models end up free/open.

  • Profilbild von Roxana Ivan

    @RoxanaIvankkea

    Another things chinese are doing is factory automation. This does not need llms but uses proven workhorses from ml/ai realm. This requires apecialized models, but for which you don't need super duper infra. There is a learning curve for making these things efficient. Way ahead

  • Profilbild von Simon Thompson@AAMAS 1999

    @AiSimonThompson

    Hang on - these guys stole the worlds intellectual property! They do not get to make this argument.