Extropic founder Guillaume Verdon says decentralized and open-source AI is regaining momentum against centralized systems

I endorse this sentiment entirely.

@beffjezos How do we fund this, though? The main advantage centralized labs have is economies of scale

Centralized models hit a wall in domains where data is sensitive, fragmented, and high-stakes (like drug discovery). The real leverage now comes from systems that can train powerful models where the data already lives, without forcing it into a single pot. Client-owned, federated approaches are of course more private but crucially they’re often higher performance because they train on the actual distribution of real-world data, not sanitized public benchmarks.

@beffjezos Bittensor

@beffjezos Yes it is

$TAO

@ChrisVolkernick @beffjezos The economy of scale for local hosted is you don't need 2T param models lol, the net demand on resources is significantly less.

@beffjezos Bittensor

Although I don't see it being very measurable, once a well needed breakthrough happens for distillation and most importantly the process of re-educating each student expert to be more intelligent than its expert teacher. Then open source local will fly. That's the next giant thing. Because most the people globally can't afford to host UN-quanted GLM5.2 today.

@beffjezos Finally someone gets it. Open source is the only way to keep things real and censorship free.

@beffjezos It will definitely be growing for sure and almost the only way for a lot of things

@beffjezos I saw that coming with forulus, a local, secure vault for personal medical data

@beffjezos Word.

@ChrisVolkernick @beffjezos $TAO is how you fund it. An incentive machine compounded by real PMF.

@BrianRoemmele No more space x stock shilling Brian?

@BrianRoemmele I hope you are right. My farm has 4 experimental growing rooms full with sensors, waiting for this happening in terms of real life practical HW. I don't want to invest now on M3 studio hw, expecting something is cooking somewhere inside Apple!?

@ChrisVolkernick @beffjezos …but worse performance and more vulnerability to disruption. Best of both worlds? Federated Learning (See: Qubigen)

@df00z @beffjezos Idk if I agree with that writ large; for the majority of things, you are probably correct. You don't need the frontier models. However, you don't get the real cutting edge capabilities on those lesser models. You need a frontier model for orchestration tasks

@beffjezos how will decentralized intelligence reshape our daily decisions?

@beffjezos the pendulum always swings back

@beffjezos mosey over to my page and you will see that Crackhead Intel is not making any comeback whatsoever, despite being right on the money 99% of the time