BitTensor übernimmt Anthropics Anordnung zur Abschaltung des KI-Modells – hier ist der Grund

Alrighty, let’s do this.

Given that TAO markets itself as „decentralized AI“, a regulatory shut down of a model such as Anthropic Mythos highlights the benefits of „decentralized AI“. That line of reasoning works if you stop your thinking there and never ask any further questions.

So, what is TAO? TAO is a middle man, a routing layer between requests and Gen AI responses. It uses crypto as an incentive model for people to put up hardware to run compute. The crypto side of things could have also been done with credit cards.

So, putting the „crypto“ element aside, how is it decentralized AI? To be clear, if Anthropic Mythos is banned, it would also not be available via TAO. Other models would be fine though, but those same models are available directly from their respective vendors (OpenAI, Google, etc).

But what about open source models such as the Llama ones, those can’t be shut down, right? Yes, that’s great! But the value there doesn’t come from TAO here, it comes from the existence of the open source models, their ongoing funding and development (Meta…), and the hardware provider that runs them.

OK, so what about hardware? Well, almost all TAO inference hardware is run of course on the big three clouds, because that is the most cost effective way to compute Gen AI workloads. But what about individual users offering their GPU? That is too slow to do any real work, and not something you would put production workloads on.

TAO *does* provide value, the idea of providing an abstraction layer behind hardware and models is neat to an extent, and there are competitors that offer that. In reality most companies choose their cloud provider and stick to them. Multicloud is real and some ultra large companies do that, and those may use a routing service, but that does not make „decentralized AI“ or shield you from regulatory risk in any real way.

You can still access anthropic models using a third party such Venice AI for instance.