When Google’s researching your business model, you’re not early anymore. You’re right.
We already have this live and working.
260,000+ phones. 175+ countries.
Today on the blog, we discuss a pathway for the second life of phones through the exploration of “phone cluster computing”, which can directly reduce the environmental footprint of computing by avoiding the need for further raw material extraction. More →goo.gle/4aJe5vO
pierreaubert retweeted
Read the paper: arxiv.org/abs/2605.27276
Github Repo: github.com/hexo-ai/sia
pierreaubert retweeted
Fast, faster, Qwen. 🚀
Thrilled to see Qwen3.5 reaching a record-breaking 580 tps for agentic workloads on the TokenSpeed engine! This milestone wouldn't be possible without our incredible partners.
Huge thanks to @lightseekorg, @NVIDIAAI, the Mooncake team, and @tri_dao for the pioneering FA4 optimization. Together, we are pushing the boundaries of open-source LLM inference. 🤝✨
Dive into the full @PyTorch blog post below! 👇
pytorch.org/blog/up-to-580tp…
#Qwen #Qwen3_5 #TokenSpeed #LLM #Inference #AI #PyTorch #OpenSource #AgenticAI #HighPerformance
The speed-of-light optimization for Qwen3.5 on the TokenSpeed inference engine is a significant milestone, achieving a record-breaking 580 tokens per second (tps) for agentic workloads on NVIDIA GPUs.
In the PyTorch Foundation's latest community blog post, you can learn all about the complete design, implementation, and optimization of Qwen3.5 models in the TokenSpeed inference framework and see for yourself how this work is improving performance 👉 bit.ly/4uGUvIS
This achievement was a joint effort between the @Alibaba_Qwen inference team, @lightseekorg Foundation TokenSpeed team, @NVIDIAAI , and the Mooncake team, with special contributions from @tri_dao for FlashAttention-4 (FA4) optimization. @KVCache_AI
pierreaubert retweeted
I built a simpler and fully decentralized implementation of the @CoinbaseDev x402 payment protocol.
No "facilitators". No RPC providers. No API keys.
Meet smol402.
pierreaubert retweeted
The Web3 Foundation wishes you a wonderful holiday season.
We appreciate the collaboration and progress made together across the ecosystem🎄
pierreaubert retweeted
1/ Concerns for AI-generated content, age verification, and money laundering are driving online ID requirements. Zk proofs about your ID are not enough to do this privately or securely. Some new ( and less new) work from my lab addresses this. eprint.iacr.org/2025/1938
pierreaubert retweeted
Anthropic is now testing on smart contract exploits and discovering zero-days.
This stuff is going from theoretical to practical scarily fast.
New on our Frontier Red Team blog: We tested whether AIs can exploit blockchain smart contracts.
In simulated testing, AI agents found $4.6M in exploits.
The research (with @MATSprogram and the Anthropic Fellows program) also developed a new benchmark: red.anthropic.com/2025/smart…
pierreaubert retweeted
I finally finished my smol block article about 500ms blocks: blog.kchr.de/from-12s-to-500…
The SRE in me is please to report some great improvement in tx in block confidence: <55% to >99% @99.9%tile 1d. Go Polkadot.
I disagree: we definitely have the talent. For leadership i will let you judge.
pierreaubert retweeted
Exclude Circle & Tether, top 8 projects = ~55.5% of revenue last yr.
500+ others share ~44.5%.
tldr:
• Nothing new. Tether & Circle = best businesses in crypto. Together, make up ($7.4B) over 52% of total revenue.
• L1s still dominate: ~1/3 of rest.
• Bots are outperforming AMMs
• SocialFi kinda flopped. Friendtech, Lens, Farcaster all generated less than a single decent meme bot. Only Farcaster remains relevant at this point.
• Infra tolls (Dexscreener, Zeebu) actually monetize decently within their niche.
Ethereum still top w/ $943M+.
TRON feasting on stablecoin txs, still defacto for that.
But the most successful products this cycle are Web3 Casinos.
And that’s exactly what the market is rewarding and optimizing for.
Pumpfun with $595M.
Hyperliquid, pulling in $174M in less than half a year.
You also see Phantom ($388M) and Photon ($345M), both outperforming nearly everything else.
New UX plays especially for better casino experience (intent-based, bundling, MEV-aware routing) eating trad AMMs.
And therefore, new builders are much more incentived to build the next better trading bot and pumpfun.
Beside that in general, wallets and infra tools are monetizing well too whether it’s routing, swaps, data, or affiliate flows.
You can now make good money just by a frontend without touching core assets.
And rounding out the rest of the list, Solana ecosystem products continue to dominate: Jupiter, BonkBot, Raydium, Kamino etc.
So it’s evident, this cycle, the market isn’t rewarding ethos, decentralization, resilience.
It’s rewarding dopamine loops, sniping flows, and speculative UX for wild desires.
Pumpfun generating more revenue than entire L1s.
Totaling nearly $600M in app revenue over the past year.
Looks like Web3 is indeed going great.
pierreaubert retweeted
Wow~@Polkadot 的又一个重大进展!💥
大家一直说 Polkadot 尽管技术很强大,但对开发者来说却很痛苦...😭
过去 6 个月,@paritytech 团队一直在专注开发 Polkadot 的“一键部署”解决方案 —— Polkadot 部署门户(PDP:Polkadot Deployment Portal)。无需复杂操作,1 秒上手,一键部署 Rollup!
现在,PDP 即将上线,让 Web3 开发者告别繁琐配置,快速部署、管理、扩展 Rollup:
✅ 零门槛体验:无需深入区块链底层,轻松选择运行环境 & Coretime
✅ 超快部署:从 几天 → 几秒,真正实现一键 Rollup
✅ 灵活配置:支持 Polkadot / Kusama / Westend / Paseo,适配不同需求
Web3 将真正进入无许可创新时代,人人都能构建自己的区块链!
🔥 现开放 3-5 名 Kusama 早期测试者!将在 2025 年 4 月中旬在 Kusama 上进行生产环境的部署!
📩 立即申请加入 :docs.google.com/forms/d/1th3…
Interested in simple rollup deployment on Polkadot? You can read about the new Polkadot Deployment Portal and even sign up to be an early tester here: forum.polkadot.network/t/pol…
pierreaubert retweeted
Start building a dApp on Polkadot in less than 5 minutes!
Our team just published the 1.0.0 of a new tool that lets you get started quickly, and without any headache.
One step closer to crafting the best DevX on @Polkadot
gg @byteboro , @mutantcornholio & team!
I am so jalous … maybe start with one? Xmas is far away :)
Running DeepSeek R1 on my desk
Uses @exolabs with Thunderbolt 5 interconnect (80Gbps) to run the full (671B, 8-bit) DeepSeek R1 distributed across 2 M3 Ultra 512GB Mac Studios (1TB total Unified Memory).
Runs at 11 tok/sec. Theoretical max is ~20 tok/sec.
Daily frustration:
python perfcli.py
Speedup for depth 17 (131071 nodes):
Threads Garble Speedup Eval Speedup Total Speedup
-------------------------------------------------------
1 1.00 1.00 1.00
4 1.00 1.01 1.00
0 speedup with multithreading, nogil, cpython, ...





























