记录今天 11/3/25，到时回看🥰

Shipped Rapid-MLX v0.7.3 — bringing best performance and structured tool calling to DiffusionGemma 26B on Apple Silicon. 🛠️ Engine optimizations pushed throughput to 50 tok/s on an M3 Ultra. Zero parsing headaches. v0.7.3 perfectly handles DiffusionGemma's tool calls natively right out of the box. My new fav local LLM now! 🔗 rapidmlx.com

Apple is the true future of decentralization. The narrative that Apple is "lagging" in the AI arms race completely misses the mark. They aren't failing to compete with hyperscalers—they are executing a profoundly different playbook. 🎯 While the industry forces us to rent intelligence from centralized monoliths (trading our privacy and paying exorbitant token fees 💸), Apple is quietly building the exact opposite: 👤 People-owned AI, running entirely on people-owned devices. 🧠 Core AI (announced in WWDC'26) transforms the device in your pocket from a mere data-harvesting terminal into a sovereign engine of your own intelligence. 🤝 It’s a massive shift: Full Privacy, Low Latency, Zero Token Fees and True AI empowerment isn't a subscription to a distant server. It's local, private, and unequivocally yours am bullish $APPL (as the developer of rapid-mlx inference engine on MacOS)

NOBODY IS READY FOR $IOTX The only question is: are you positioned?

That face you make when you realize you’ve been overpaying AI Bill Stop letting overpriced AI drain your wallet. QuickSilver Pro is officially the cheapest API for top-tier open-source LLMs. We’re 20% cheaper than OpenRouter and up to 75% cheaper than Together AI & Fireworks. Same models, same SDK, way more savings. $IOTX Visit quicksilverpro.io

🔴 The Pain: Running local MLX models is incredibly fast and private. But let's be real - testing tool calling via terminal is clunky, and there's zero good UI for it. 🟢 The Fix: rapid-mlx share Just ONE command gives you a polished web chat seamless tool calling (works beautifully with gemma-4-12b-qat). We are proud to be the ONLY MLX inference engine in the community shipping this. ⚡️ 👇 Try it now: brew install raullenchai/rapid-mlx/rapid-mlx

Physical Intelligence requires a frictionless user experience. With EIP-7702 arriving, Your regular wallet just gained smart contract powers. 🪡🧵 $IOTX

YOU BETTER GET SOME $IOTX EXPOSURE BEFORE ITS TOO LATE NOW

Your OpenAI-compatible API makes images too 🎨 QuickSilver Pro now serves FLUX.2 Pro & FLUX.1 Schnell — same key, same SDK, same bill you already use for chat. ✅ Zero code rewrites — just client.images.generate(model="flux.2-pro", …) ✅ One key, one bill for text and images ✅ Per-image pricing ($0.027 / $0.0025), usage.cost on every call Ship it today 👇 quicksilverpro.io/docs/image…

Quantum computing isn't a 'maybe', it's a 'when'. I'm betting big on the quantum future, which means future-proofing @iotex_io today. IIP-64 is how we build the ultimate shield for @iotex_io mainnet.

IIP-64 is open for discussion: Post-Quantum Cryptography. 🛡️ Following recent quantum breakthroughs, we're evolving IoTeX's core cryptography from "classical ECDSA" to secure IoTeX against next-gen quantum threats. Here’s the proposal: 🛡️Post-Quantum Security — Migrating from secp256k1 to ML-DSA-65 ⚡Tailored for IoTeX — Optimized for 5s block finality and Roll-DPoS consensus. 📈 Proactive Defense — Aligning with NIST standards to secure user assets。 Read the full proposal and share your thoughts👇

Real-world AI doesn't happen overnight. We’ve been quietly building behind the scenes, and QuickSilver Pro is just a glimpse of what's coming. The foundation is set. Big things are ahead. 🚀🔥 Join 1,000 developers building now: quicksilverpro.io

Real-World AI 并非一蹴而就。QuickSilver Pro 只是我们未来宏伟 AI 蓝图的冰山一角。 立即加入超 1,000 名开发者的行列，共同构建未来：quicksilverpro.io

"AI is plateauing" is the laziest take of 2026. 🥱 I mapped ~110 major LLM releases across 11 labs over the last 30 months. We aren't slowing down; we're mutating. 🧬 ▪️ 2024: 35 releases, 6 paradigm shifts ▪️ 2025: 36 releases, 4 paradigm shifts ▪️2026 YTD: On pace to shatter both records Every ~9 months, a completely new dimension unlocks: 🧠 Sept '24 — Inference-time compute (o1) 📉 Dec '24 — Training cost collapse (DeepSeek V3, ~$6M) 🔓 Jan '25 — Open reasoning (R1) ⚡ Jan '26 — Linear attention at frontier scale (Qwen 3.5, DeltaNet) The frontier has consolidated to ~5 apex labs 🦅—and they are shipping faster than 50 did in 2020. The singularity isn't a smooth curve; it's a staircase. 🪜

数字资产监管的里程碑时刻！ “CLARITY 法案”已正式通过参议院银行委员会表决，这标志着美国在建立清晰的数字资产框架进程中迈出了关键一步。 无论是现实世界资产 RWA 的代币化，还是AI驱动的经济形态，若要实现全球范围内的规模化扩张，法治的确定性都是不可或缺的底层支撑。 随着该法案挺进参议院全体审议，支撑数字经济未来的行业基石正加速成型。

A landmark moment for digital asset regulation. The CLARITY Act has cleared the Senate Banking Committee, moving the U.S. closer to a clear framework for digital assets. Real-world tokenization and AI-driven economies require a foundation of legal certainty to reach global scale. With this bill heading to the full Senate, that foundation is finally being built. 🇺🇸

在 Apple Silicon 上以高达 ~2 倍的速度运行 Qwen 大模型 无损且单流： Rapid-MLX v0.6.37 现已为 Qwen3.5/3.6-27B-8bit 默认启用 DFlash 推测性解码 (Speculative Decoding) 通过 mlx-vlm 实现的 z-lab Block-diffusion Drafter：在单次前向传播中起草一整块 Token，然后进行并行验证。其输出分布与自回归模式完全一致 —— 在大多数工作负载下均通过了位级 (Bitwise) 验证。 Qwen3.6-27B-8bit (平均提速 1.49×): 2.07× 摘要生成 1.82× 代码编写 1.70× 数学运算 1.22× 闲聊 1.08× 长文本生成 1.06× 智能体任务 ✓ 无单项工作负载性能回退 Qwen3.5-27B-8bit (平均提速 1.31×): 2.15× 代码编写 1.66× 数学运算 1.47× 摘要生成 1.18× 智能体任务 0.80× 闲聊 0.59× 长文本生成 立即体验： pip install 'rapid-mlx[dflash]' rapid-mlx serve qwen3.5-27b-8bit --enable-dflash

🚀 Run Qwen LLMs up to ~2× faster on Apple Silicon. Lossless & single-stream. ⚡ Rapid-MLX v0.6.37 now enables DFlash speculative decoding by default for Qwen3.5/3.6-27B-8bit. z-lab's block-diffusion drafter via mlx-vlm: drafts a whole block of tokens in ONE forward pass, then verifies them in parallel. Same output distribution as autoregressive — verified bitwise on most workloads Qwen3.6-27B-8bit (Avg 1.49×): 🟢 2.07× summarize 🟢 1.82× code 🟢 1.70× math 🟢 1.22× chat 🟢 1.08× longgen 🟢 1.06× agent ✓ No per-workload regression Qwen3.5-27B-8bit (Avg 1.31×): 🟢 2.15× code 🟢 1.66× math 🟢 1.47× summarize 🟢 1.18× agent 🟡 0.80× chat 🔴 0.59× longgen Try it: pip install 'rapid-mlx[dflash]' rapid-mlx serve qwen3.5-27b-8bit --enable-dflash

Anthropic just quietly dropped a nuke on Wall Street workflows. 🤯 They open-sourced a massive Agentic AI toolkit for the financial industry. If you build AI agents or work in finance, drop everything and look at this repo. The anthropics/financial-services repo provides out-of-the-box Claude Agents, specialized skills, and MCP data connectors (S&P Global, FactSet, PitchBook) designed for high-stakes workflows. Why is this a big deal? It fundamentally shifts AI from "glorified chat" to actual Agentic automation. Instead of just answering questions, these agents execute complex, multi-step tasks, e.g., 1️⃣ The "Pitch Agent" - Instead of an analyst grinding for 10 hours overnight, this agent pulls live financial data, runs an LBO or DCF model, and auto-generates a branded PPT pitch deck in minutes. It turns data gathering and formatting into a one-click operation. 2️⃣ The "Earnings Reviewer" - Imagine a stock just reported earnings. The agent instantly ingests the live call transcript, compares it against historical Morningstar estimates, automatically updates your Excel model with new EPS guidance, and drafts a morning research note before the market even opens. 3️⃣ The "Deal Screener" - Reviewing endless inbound pitch decks and teasers is exhausting. This agent can automatically parse incoming Confidential Information Memorandums (CIMs), run a quick pass/fail assessment based on your specific investment thesis, and even draft the initial IC memo. Link in the first reply 🧵👇

@iotex_io v2.4.0 is shipping — our biggest mainnet upgrade of 2026. It's a mandatory hardfork activating at block 48,985,561 (≈ June 7, 2026 · 23:52 UTC). Two things we've been working toward all year: ⚡ Pectra EVM — IoTeX is now aligned with Ethereum's latest execution layer: · EIP-7702 — EOAs can act as smart accounts · EIP-2537 — native BLS12-381 (bridges, light clients, ZK) · EIP-2935 — historical block hashes from state · EIP-7623 — recalibrated calldata pricing 🛡️ Candidate Exit Queue — delegates can no longer unwind their self-stake instantly. Exits now go through a request → schedule → confirm flow, with admission rate-limited at the protocol level. A cleaner, more predictable lifecycle for our delegate set. What you need to do 👇 ▸ Delegates / Full Nodes / API Nodes — upgrade to v2.4.0 before block 48,985,561. Miss it = fork off mainnet. ▸ Wallets & dApps — check your gas estimator against the new EIP-7623 calldata floor, and add support for TxType 0x04 (SetCode) if you serialize raw transactions. ▸ Indexers / Explorers — new on-chain events from the exit queue and 7702 authorization tuples are ready to be indexed. Release notes: github.com/iotexproject/iote… Huge thanks to the core team, delegates, and every dev who helped pressure-test this release. Run a node. Build on it. See you on the other side of 48,985,561. LFG. 🟢

We are coming back strong, in every way!