推文
100 tweets · 45 sources账号:全部mattpocockukbchernydair_aifelixriesebergtrq212simonw0xblacklightadocompletedexhorthydani_avila7ryancarsonGeminiAppgarrytankarpathymckaywrigleynummanaliomarsar00xPauliusAnthropicAIDanielMiesslerGoogleDeepMindRaillyHugoaidenybaiasynkimocgtwtsctatedevdemishassabisdillon_mulroydoodlesteinelithrarjarredsumnerjoshalbrechtjpschroederlennysanlexruslogancyangpetergyangrohanpaul_aisawyerhoodservasyy_aithdxrthorstenballycombinatoryiliushzarazhangrui
@simonw·6d ago
I've been vibe coding SwiftUI menu bar apps for my new Mac, turns out Claude Opus 4.6 and GPT-5.4 are both competent at Swift programming, no need to even open Xcode! https://t.co/1Vi0rtlVh1
我一直在用 vibe coding 的方式为我的新 Mac 开发 SwiftUI 菜单栏应用，结果发现 Claude Opus 4.6 和 GPT-5.4 在 Swift 编程上都很能打，甚至不需要打开 Xcode！
@simonw·3月26日
To me this mostly illustrates the futility of robust jailbreaking prevention
@kotekjedi_ml
New paper: We deploy Claude Code in an autoresearch loop to discover novel jailbreaking algorithms – and it works. It beats 30+ existing GCG-like attacks (with AutoML hyperparameter tuning) This is a strong sign that incremental safety and security research can now be automated. https://t.co/cDwxVydVPr
@simonw：对我来说，这主要说明了强健 jailbreak 防御的徒劳性。 [引用 @kotekjedi_ml]：新论文：我们将 Claude Code 部署在自动研究循环中，用于发现新型 jailbreak 算法——而且效果显著。它击败了 30 多种现有的 GCG 类攻击（配合 AutoML 超参数调优）。 这是一个强烈信号：增量式安全研究现在可以被自动化了。
@simonw·3月24日
Thankfully the LiteLLM package has now been marked as "quarantined" on PyPI so attempting to install the compromised update via pip et al shouldn't work https://t.co/BmrbWCoLXn
@hnykda
LiteLLM HAS BEEN COMPROMISED, DO NOT UPDATE. We just discovered that LiteLLM pypi release 1.82.8. It has been compromised, it contains litellm_init.pth with base64 encoded instructions to send all the credentials it can find to remote server + self-replicate. link below
好消息：LiteLLM 包已在 PyPI 上被标记为「隔离」，因此通过 pip 等工具安装被攻击的更新版本应该已经无法成功。 > 【引用】LiteLLM 已被入侵，请勿更新。我们刚刚发现 LiteLLM pypi 1.82.8 版本遭到投毒，其中包含 litellm_init.pth 文件，内含 base64 编码的指令，会将所能找到的所有凭证发送至远程服务器，并进行自我复制。
@simonw·3月24日
Turns out you can run enormous Mixture-of-Experts on Mac hardware without fitting the whole model in RAM by streaming a subset of expert weights from SSD for each generated token - and people keep finding ways to run bigger models
Kimi 2.5 is 1T, but only 32B active so fits 96GB
@seikixtc
I got a 1T-parameter model running locally on my MacBook Pro. LLM: Kimi K2.5 1,026,408,232,448 params (~1.026T) Hardware: M2 Max MacBook Pro (2023) w/ 96GB unified memory Running on MLX with a flash-style SSD streaming path + local patching. This is an experimental setup and I haven’t optimized speed yet, but it’s stable enough that I’ve started testing it in an autoresearch-style loop. #LocalAI #MLX #MoE
原来可以在 Mac 硬件上运行超大型 Mixture-of-Experts 模型，而无需将整个模型载入内存——方法是在每次生成 token 时从 SSD 流式加载一部分专家权重。而且人们还在不断想办法跑更大的模型。 Kimi 2.5 有 1T 参数，但每次只激活 32B，所以能放进 96GB 内存。 [引用 @seikixtc]：我在 MacBook Pro 上跑起来了一个 1T 参数的模型。 LLM：Kimi K2.5 参数量：1,026,408,232,448（约 1.026T） 硬件：M2 Max MacBook Pro（2023），96GB 统一内存 使用 MLX 运行，采用 flash-style SSD streaming 路径 + local patching。 这是实验性配置，速度还没有优化，但已经稳定到足以在 autoresearch 风格的循环中测试使用。 #LocalAI #MLX #MoE
@simonw·3月23日
Starlette 1.0 is out! I used this as an opportunity to experiment with Claude Skills, since Claude isn't yet familiar with the (minor) breaking changes in the 1.0 release compared to 0.x https://t.co/IxQ18DShxg
Starlette 1.0 发布了！我借这个机会尝试了 Claude Skills，因为 Claude 对 1.0 版本相比 0.x 的（小）breaking changes 还不熟悉。
100 tweets · 45 sources