Building the infrastructure for decentralized intelligence.
Joined December 2023
- Tweets719
- Following373
- Followers2.4K
- Likes624
Bittensor Weekly Roundup
Quantum-resistant wallets head for Release 487, Finney moves to runtime 486, an audit fix batch goes up for review, Targon burns SN4 alpha from revenue, and lium and Chutes ship new compute and private inference.
5. UPDATE from @lium_io SN51: lium now has 550 B300s, with 443 of them rented, and is pricing B300s at $4/hour, the same as H200s.
teapaw.org/lium_io/status/2115445…
6. NEW from @chutes_ai SN64: SayGM private models can now run on Chutes inside secure hardware, where Chutes says no one can read prompts, not even Chutes. Subnet stacking, SN28 on SN64.
teapaw.org/chutes_ai/status/21154…
Chutes is now live on @say_gm_
SayGM gives you one API key for Claude, GPT, Gemini and open-source models.
When you pick one of SayGM’s private models, it can now run on Chutes, inside secure hardware where no one can read your prompts. Not even us.
See the full article here: chutes.ai/news/chutes-models…
Great product release from @TPN_Labs 👏
Introducing gratis-ai.com
Run open source AI models locally in your browser, in just a few clicks.
Pick an opensource model or load any model from @huggingface. It downloads to your machine and runs there privately; every prompt stays on your device.
Get your free personal AI today.
AI agent adoption at scale requires trust, not just intelligence.
Making SOTA security cheaper, smaller and practical enough to check every output is a major step towards production-ready agents.
Another example of what open competition on Bittensor can deliver. Ψ
Every company shipping AI agents asks one question: how can I stop my agent from going rogue?
To answer it, HaloGuard Output Guard now sets a new state of the art (SOTA):
83.63 average F1, ahead of the previous best at 82.5. All measured on public benchmarks.
And we did it with a 0.8B model, a quarter of the size of the previous best. Small enough to run efficiently. Cheap enough to make checking every response practical.
This was the focus of Phase 3 on Trishool:
• 7,500+ jailbreaks submitted
• 125+ unique miners
• 15 weeks of challenges
It comes at the right time. We’re taking our product to enterprises now, and this makes our security stack stronger.
We’ll share the models on Hugging Face, the arXiv paper and a full explainer soon. There’s much more to come, and we’re looking forward to an exciting week ahead.
SOTA AI security, on Bittensor. Ψ
5. NEW: Ingot (SN97) is a new model network. Its latest release, Ingot-I, quantized Qwen 27B to FP8 better than Alibaba's own quantization in one week, per @const_reborn. Next targets: 2-bit, then larger GLM models.
teapaw.org/const_reborn/status/21…
6. UPDATE: Score (SN44) set a 90% target for fallen-person detection on crowded street footage. Miners hit 98%, up from a 43.6% baseline, scored publicly every hour with open weights. Harder footage goes in next.
teapaw.org/webuildscore/status/21…
3. UPDATE: IOTA (SN9) cut merge time in decentralized training about 5x, making training almost 20% cheaper. The fix: turning the network graph into a tree during merges, so the swarm spends less time paused waiting on its slowest member.
teapaw.org/IOTA_SN9/status/211225…
4. NEW: REAX (SN92) released REAX 1, an open decision model on Perplexity Decider v1.2. The team measures +2.4 points over its base across all 6 test draws, and opened a miner competition to beat it.
teapaw.org/reaxlabs/status/211286…
Meet REAX 1, our open decision model built on Perplexity Decider v1.2.
Our own measurement: +2.4 points over its base, ahead in all 6 test draws and all 10 areas.
Open competition starts today 13:00 UTC:
=> Beat it, earn the rewards
see reax.co/competitions
2. UPDATE from @opentensor: Runtime 482 shipped. Coldkey swaps can now merge into an already-used wallet, and a new V2 announce call reserves the destination during the delay. Also new Validate and Weights proxy types for hotkeys.
github.com/RaoFoundation/sub…
1. NEW from @opentensor: eTAO is live with Runtime 481. Subnet owners can now set their own alpha emission split: owner 0 to 33%, validators up to 50%, miners at least 33%, changeable once a week. Default stays 18/41/41.
teapaw.org/opentensor/status/2112…
The tradeoff: miners are now guaranteed 33%, down from a fixed 41%, so owners can shift weight to validators or themselves. Root stakers are fenced off: the chain pays their baseline share off the top every block.
github.com/RaoFoundation/sub…
Bittensor Weekly Roundup
eTAO goes live, Runtime 482 reworks coldkey swaps, IOTA speeds up decentralized training, and REAX, Ingot and Score post new miner results.
General Tensor retweeted
Months of work. Now live.
We've launched HaloGuard EU, our first product. Built to block the AI output the EU AI Act bans or treats as high-risk. $30 / billion tokens.
Priced, live, free trial open. And it’s only the first.
