The super power of Open Source is that people can build on one anotherβs work
It is NOT ability to "run locally"
That's a nice feature, but you've been able to do that with Closed Source software from the beginning!
Ability to build on the work of others is what makes Open Source BETTER than closed could ever be
Without this ability, Open Source cannot compete..
and if you can't compete, you're irrelevant
But you canβt meaningfully build on other's work if all you have is model weights
Open Source needs to operate on the layer beneath: ALGORITHMS
Keeping people fixated on models is like keeping the masses fed while denying them the ability to grow their own food
That's a dangerous situation
Hey @VitalikButerin
You know you can start the ETH run with just doing 1 thing right?
Donate the sold 114k usdc of @AsteroidCoinOG
After that boom!
ALT seasons starts.
You know it.
From July 8, using Claude can require a scan of your government ID and a biometric map of your face.
Anthropic is (partly) doing this because its model is being stolen.
This week it accused Alibaba of running 28.8 million queries through Claude to train a rival, the largest extraction campaign it has ever caught.
We spoke about these distillation attacks before (https://t.co/5QZa1TWxdf) and how they are the primary reason open source models are competitive.
No lab is going to spend billions training a model and let that value be stolen.
Once they stop these attacks, open-source models will increasingly lag behind.
One of the ways they will stop this is by requiring your identity.
If you are suspected of taking part (e.g. selling your subscription, as in the pic below), you will be permanently banned.
Anthropic, OpenAI and Google already share one channel Frontier Model Forum, for pooling intelligence on the extraction.
So you could be banned from ALL frontier models.
And soon, the bar for requesting ID and enforcing bans will lower.
You could get banned for trying to use Fable to improve an AI model. Or asking it a question it doesn't like. That ban carries across ALL labs, and you lose frontier access entirely...
....a loss so big you would not risk it in the first place.ξξ»ξ
ξ»ξΉξ
Long before they ban anyone, you start policing yourself.
Frontier AI began as something anyone could reach.
It is ending as something you qualify for.
We HAVE to ensure models stay open.
Keep the algorithms open and you keep intelligence open.