/Tech6h ago

Garry Tan argues open-source AI is necessary for enterprises to avoid losing proprietary training data to major AI labs

Labs use customer workflow data to refine reinforcement learning

472982711250.1K

#121

Original post

Garry Tan@garrytan#121inTech

Open source is the escape hatch for businesses to be able to continue to control their own destiny long term

dom@ryiacy

My interpretation of this:

Right now, Anthropic and OpenAI are making a killing by selling enterprise FDE services to F500s, building workflows for them on top of proprietary models, then using the traces and context from this to build RL envs to improve the models.

This is crazy amounts of leverage - instead of buying this data they're getting paid gigantic consulting fees to extract it.

This also goes way beyond typical consulting in scope - organizations are effectively outsourcing key learning curves and domain knowledge to the AI labs.

Despite that, it's so far been worth it for them because the value of skilled FDE is so high and the ROI so fast, and orgs are willing to pay a premium for competent AI implementation.

But in the long run, one of two things happens: either orgs are gonna get hooked on this and end up paying for the model training that replaces their business, or they find a way to build and own their own model ecosystem.

What that looks like is developing some combination of AI models, evals, RL envs, and workflows. Initially probably the model will still be an off-the-shelf frontier model from a top lab.

But as firms build out more sophisticated eval / RL env (increasingly the same thing) infra, it starts to become viable to post-train an custom model on top of an OSS base. Cursor have done this successfully with their Composer model RL'd on top of Kimi.

Sidenote, this is the same conversation that a lot of national governments in Europe are having in the past week. When we look at what the rhetoric about 'sovereign AI' in the UK actually boils down to, it's doing custom post-training on top of an OSS model, and then running it on local GPUs.

Ultimately, the current feeding frenzy for AI services in all of its guises - FDE, AI consulting, etc - should raise questions about long-term sustainability. If consulting services are truly a value add and competitive advantage, then in the long term you want to in-house.

4:51 PM · Jun 14, 2026 · 49.4K Views

Sentiment

Many users view open source as the key escape from vendor lock-in with Anthropic and OpenAI enterprise offerings, while others argue no genuine open source exists yet and both companies' models fall short.

Pos

63.2%

Neg

36.8%

20 comments with sentiment.

Cluster Engagement

Posts from X

Most Activity

VIEWS1.2KBOOKMARKS1RETWEETS1REPLIES1

Andrew Carr 🤸@andrew_n_carr

Open question. Could you post train a fable level model on your specific task while starting from an open base model?

dom@ryiacy

My interpretation of this:

This is crazy amounts of leverage - instead of buying this data they're getting paid gigantic consulting fees to extract it.

This also goes way beyond typical consulting in scope - organizations are effectively outsourcing key learning curves and domain knowledge to the AI labs.

Despite that, it's so far been worth it for them because the value of skilled FDE is so high and the ROI so fast, and orgs are willing to pay a premium for competent AI implementation.

What that looks like is developing some combination of AI models, evals, RL envs, and workflows. Initially probably the model will still be an off-the-shelf frontier model from a top lab.

2h1.2K41

LIKES22

Kevin@kevincodex

@garrytan that is why @gitlawb is open source and our thesis revolves around it. open source will win

6h28722

Aurea@AureaLibe

As it has always been. Little by little, we’re running models on smaller and smaller hardware. companies will be able to train their own custom models tailored to their core business without having to deal with large, closed, and unnecessary models.

On top of that, it allows them to own their data and reduce their costs, because there’s a token spending ceiling they can afford without hurting profitability. Once you cross that threshold, it can become critical. Open source fixes this.

6h8531

Behrad Khodayar@behradkhodayar

@garrytan Europe r u listening?

6h461

Chris Nota@chris_nota_rl

@behradkhodayar @garrytan If the cat model is real, then they might be after all!

6h111

Donald Tech@wajalein

@garrytan In China, every company must have a Party branch, and there are even policies now that strictly prohibit technical personnel from leaving China. Despite this, I am still trying to find a solution!

6h1392

Sean D Langshaw@sdlangshaw

@garrytan Microsoft has entered the fray.

They already have the F500 CTO contracts, the govCloud, sovereign Clouds.

It's only a matter of time for model providers.

Building model agnostic, observability, platform agnostic, with governance.

Opensource comes after winning contracts.

5h1921

Ericlamideas@ericlamideas

@garrytan all serious open source is chinese, US companies can't leverage b/c strict data residency rules. we need more US open source

5h721

🏆🇧🇷💎Jack Hunter 💎 🇧🇷Ⓜ️⚛️🎖️🥇💫@GemsHunting_8

@garrytan AI infrastructure needs to be open source and decentralized

Or forget about it!

6h721

Kenneth Eversole@kennetheversole

@garrytan We need to invest and commoditize the language model layer as fast as we can.

The best context is the most private context, and for us to control our own destinies, businesses should be empowered to keep it.

4h471

Shashank Trivedi@shashank__says

@garrytan yes it is

6h381

Matt Bjornson@mattsbjornson

@garrytan For F500 it’s not about controlling their destiny, it’s another variation of “no one ever got fired for choosing IBM”… it’s cya, its risk aversion, its protect my cushy job because I’m over my head.

4h90

Amart (LOOP)⚡️@LoopOnChain

@garrytan Do you genuinely think that open source will compete with the big labs?

I kind of view this as crypto competing with Goldman Sachs.

I think AI fundamentally changes this because it enables a much larger percentage of humanity to build.

4h81

am@am1728ff

@garrytan Open weight* Nobody is publishing their evals, envs, training code or data pipelines. There currently is no open source!

4h72

Chris Nota@chris_nota_rl

@garrytan We need frontier-scale training with open-source datasets and RL envs.

That’s a great idea actually, can I have like $1 billion?

6h54

Ryan@Ryan_liberricky

@garrytan Open source gives control but also a maintenance tax. Both matter long term

6h51

blake@_buggles

@garrytan doesnt this then depend on where the open source is running? are we going to go back to on prem hardware running open source tuned models to keep it sovereign

5h43

Ferbin@Ferbin08

@garrytan You control the code, sure. But can you actually fork and maintain it independently? Most teams can't. So you're still locked into the project's roadmap. Open source is an escape hatch only if you have the capacity.

5h39

M!Dξ.⚛️🈺@mideenigmA

@garrytan what stack will be the best for this - especially the infrastructure- hardware

6h33

Friendly Neighborhood Shaman@kaelcorwin

@garrytan for me it's a way to show my peers i can program & have skills which will make them respect me so they dont beat me up

6h26