Bittensor (TAO)

The @opentensor Foundation has long run the free RPC service that the Bittensor ecosystem depends on. It's time for SN19 to take that on.

We're proposing a redesigned free tier - no account required, no API key, rate-limited by IP.

Powered by miners who lock up alpha as conviction.

The mechanism is elegant: to serve free traffic, miners stake skin in the game. Spamming the free tier for incentive and dumping means dumping on yourself. Bittensor gets a free & reliable RPC, run by its own community.

We're putting this to an owner vote.

Vote here 👉 https://t.co/AABDFyy8di

Spent the morning going through the @verathos_ai docs and their X before starting my day.

Here are a few things that stood out to me.

The untrusted compute problem has been one of the biggest challenges for decentralised AI.

Anyone can say they ran your model on their GPU, but how do you actually verify that the computation happened the way they claim? You either have to blindly trust unknown operators or depend on expensive trusted hardware, which goes against the whole idea of decentralisation.

Verathos on Bittensor SN96 is one of the first approaches I’ve seen that genuinely tries to solve this problem.

Instead of relying on trust, they commit model weights into Merkle trees on-chain. When a miner runs inference, a cryptographic proof is generated for the exact computation that took place.

The interesting part:

• The proof system runs directly inside production vLLM during CUDA graph execution, keeping the overhead in the single-digit percentage range instead of becoming a slow layer added afterwar

Hash Rate - Ep. 177: Bitsec SN60 Beats Mythos(!)
🧙 Guest: @yubrew of @bitsecai

A Bittensor $TAO Subnet Just Beat Anthropic's MYTHOS at Cybersecurity

00:00 Introduction to Bitsec
03:01 Jailbreaking Fable: Mythos Unleashed
12:02 The Competitive Landscape
17:49 The Future of Bitsec
30:03 John Yu's Journey into Cybersecurity
40:46 Subnet Token
44:53 Sales Strategy
49:40 Final Thoughts and Future Directions