百年 AI×出海

额，所以，fable 不限时了？ 什么时候都能用了？ 和雷布斯学的耍猴营销吧

@xchase173294 Honestly, making robots feel approachable matters more than making them look human. This design direction feels much more natural for the home.

Okay but can we talk about how refreshing it is to see a robot that isn't obsessed with looking human? I came across this award announcement for the A' Design Award 2026 and there's this robot called PrimeBOT Q1 that took Gold in Personal Robotics. And the whole philosophy behind it is literally "For people, unlike people." Like, they're not even trying to replicate humans — they're rejecting it on purpose. No synthetic faces, no fake skin textures. Just minimalist geometry and soft, warm materials that feel approachable. The more I read about it, the more I think this is actually the smarter approach. We've been so obsessed with making robots look like us that we forgot to make them feel good to be around. Q1 is designed to blend into your home, not dominate it. 80cm tall — right at that social distance where it feels present but not overwhelming. I don't know about you, but I'm genuinely excited to see where this goes. This feels like a real shift in how we think about machines in our personal spaces. #PrimeBOT #Q1 #DesignPhilosophy #HumanCenteredDesign #ADesignAward

@ma_zhenyuan @Primebotglobal Honestly, this is the first home robot that doesn’t feel like a gadget or a toy. The size and design actually make sense for a living space.

I've always been kinda skeptical about having a robot at home. Not in a "they're gonna take over" way, more like… do I really want another glowing, beeping, plasticky thing cluttering up my space? @Primebotglobal Then I saw this post about PrimeBOT Q1 winning Gold at the A' Design Award in Milan and something clicked. It's 80cm tall — which apparently is the exact height for "intimate social distance." And I know that sounds like design-speak but honestly? It makes sense. It's not looming over you like some industrial machine, and it's not so small that it feels like a toy. It's just… there. Comfortably. They also designed it with soft, warm materials and smooth curves so it feels safe, not cold or mechanical. And apparently, when it's not moving, it blends into the room instead of screaming "LOOK AT ME I'M A ROBOT." I don't know if I'd call myself a believer yet, but I'm definitely paying attention now. This isn't the robot I thought I'd ever want at home — and that's exactly why it works. #PrimeBOT #Q1 #RobotDesign #HomeTech #ADesignAward #PersonalRobot

@JackQi82772 so cool!

I think I just stumbled onto something that's going to be huge. Was browsing the A' Design Award 2026 results and noticed a category I've never really paid attention to before — Personal Robotics. And the first-ever Gold winner is this robot called PrimeBOT Q1. What made me stop and actually dig deeper is that it's not trying to be a humanoid. It's not trying to replace anyone. It's designed to be a companion that fits into your life — quietly, warmly, without making a big deal out of itself. The whole thing is built around this philosophy: "For people, unlike people." And I think that's actually the smartest positioning I've seen in this space. We've been trying to make robots that look and act like us for years, and honestly? It's been kind of creepy. Q1 goes the opposite direction — minimal, approachable, warm to the touch, and open to customization. It's not loud or theatrical in its movements either — just fluid and restrained. It feels like the first time someone actually thought about how a robot should make you *feel*, not just what it can *do*. This just won Gold in Milan, and I have a feeling we're going to be hearing a lot more about it. #PrimeBOT #Q1 #PersonalRobot #FutureTech #DesignForHumans #ADesignAward

@PandaMing88 The open-source, 3D-printable shell is what makes PrimeBOT Q1 genuinely interesting. It turns a home robot from a finished product into a customizable platform—something users can shape around their own needs, taste, and imagination.

Alright I need to share this because my mind is genuinely blown. I was reading about the A' Design Award 2026 winners — just doing my usual rounds of checking what's happening in the design world — and I stumbled across this Gold winner called PrimeBOT Q1. At first I thought it was just another sleek robot concept. But then I got to the part about its outer shell being fully open-source and 3D-printable. Wait, what? You mean I could literally design and print my own custom shell for it? Swap out modules depending on what I need that day? That turns the whole thing from a product into a creative canvas. It's like the difference between buying a pre-built PC and building your own — except it's a robot that lives in your home and becomes part of your daily life.@Primebotglobal And it just won Gold at A' Design Award in Milan. The recognition makes total sense to me now — because the design isn't just about how it looks. It's about how it invites participation. That's what makes it personal. #PrimeBOT #Q1 #OpenDesign #3DPrinting #ModularDesign #ADesignAward

@LunaAI519 PrimeBOT Q1 feels less like a gadget and more like the beginning of a new product category: a robot designed to belong in the home, not imitate humans. The open-source shell is the real breakthrough.

I was just scrolling through my feed this morning and stumbled across the A' Design Award 2026 results — and apparently there's this new category called "Personal Robot" that I've never even heard of before. And the first-ever Gold in that category went to something called PrimeBOT Q1. I had to click. And honestly? I wasn't prepared for what I saw. It's not trying to look like a human. No face, no skin, none of that uncanny valley stuff that creeps me out. It's just this clean, minimalist, 80cm-tall companion with soft curves and warm materials. It actually looks like something I'd want to have sitting in my living room — which is not something I ever thought I'd say about a robot. What really got me though is that the shell is fully open-source. You can 3D-print your own mods, swap out parts, make it yours. That's not a product — that's a platform. And it just won Gold in Milan. I feel like I'm watching the early days of something that's actually going to matter. #PrimeBOT #Q1 #ADesignAward #PersonalRobot #RobotDesign #TechInnovation

@HoodyLiu @Primebotglobal so cool! seems like just the future!

I was scrolling through the A' Design Award 2026 results and honestly, a lot of the winners are stunning — but none of them made me stop and think the way PrimeBOT Q1 did. @Primebotglobal Here's the thing about design awards: they often go to things that look amazing in galleries but would feel completely out of place in an actual home. Or they reward technical complexity over emotional intelligence. Q1 seems to challenge both those assumptions. The design is deceptively simple. Minimal geometry. Soft curves. Warm, low-saturation materials. No human features, no attempt to mimic biology. And yet it feels human-centered in a way that most robots don't. The 80cm height is chosen specifically for the social distance we naturally share. Like, someone actually thought about how close you'd want this thing to be to you. It blends into the background when idle. It moves gently when active. And you can customize it through 3D-printed shells so it actually reflects who you are. That's not just good design. That's design that understands people. This Gold feels earned. #PrimeBOT #Q1 #DesignAward #ADesignAward #PersonalRobot #RoboticsDesign

codex 又双叒叕重置了！ 这周官方总共声明了七次重置。。。 真下血本了啊🤣

感觉这个问题很有意思诶 不过作者目前只是根据软件模拟发现的，棋盘越大，平衡开局越难找，和棋率也越低，目前没有纯数学证明思路 可惜自己数论学的不行。。。 不过确实是让人眼前一亮的题目啊

让 Claude 替你看视频这件事，真有人做成了插件。你是不是也攒了一堆两小时的播客和发布会录像，在收藏夹里放了几个月都没点开过？claude-video 给 Claude Code 加了一个 /watch 命令，把链接发进去就能直接问视频内容。 我读了一遍它的处理流程，省 token 的设计贯穿每一步： 有字幕先抓字幕，能不下视频就不下，需要画面时只下载对应片段 ffmpeg 抽关键帧，没字幕的视频走 Whisper 转写兜底 --detail 四档可选，从纯字幕到高消耗的逐帧精读，预算自己控制 字幕能解决的就不碰画面，画面必须看的才花钱抽帧，这个优先级排得很合理。依赖 yt-dlp 和 ffmpeg，第一次跑自动装，MIT 协议。装法是在 Claude Code 里执行 /plugin marketplace add 加仓库名，一分钟就能装完。 GitHub：github.com/bradautomates/… 对要精读发布会、整理播客要点的内容创作者，这个比自己截图喂模型省事太多。字幕优先的设计决定了它处理访谈类内容效果最好，纯画面演示类的视频效果怎么样，还得等更多人用过才知道。

中国模型吃下美企三到四成 token！ 这条曲线陡得反常：一年前中国模型在美国企业网关的 token 占比只有 4.5%，CNBC 这周的调查说现在是 30% 到 46%。OpenRouter 上从 2 月 8 号那周起就没下过 30%，峰值到了 46%，而之前十二个月的均值才 11%。从个位数涨到接近一半，只用了一年出头。 驱动力只有价格一项，OpenRouter 的人给的口径是中国开源模型比一线闭源便宜 60% 到 90%。最典型的是 GLM-5.2：SWE-bench Pro 跑 62.1%，比 GPT-5.5 的 58.6% 还高，价格却只要 1.4 美元输入、4.4 美元输出，MIT 协议里明写不设地区限制。 Vercel 向 CNBC 确认，GLM-5.2 上线第一周日 token 量翻了 27 倍，客户数翻了 80 倍。美国公司一边讲安全审查，一边照常采购，这两件事看来完全可以同时发生，毕竟采购部门看的是发票，安全团队看的是清单，两边互不干扰。 剩下的悬念是这个占比的上限在哪：46% 已经见顶，还是会继续往上走？

@AomyYing 确实这段时间比较忙。。。

@bainianAI 我去 好久没刷到你了 🥹

@FinnTsai88 手机版本确实会舒服很多

@Bearlagcy cool，AI真的改变一切了

agent 跑长任务最贵的一笔开销是记忆：上下文越积越大，token 消耗越来越快，模型的输出质量反而在下降。腾讯云开源的这个记忆层，官方自测把检索任务的 token 用量降了 61.38%。 我翻了 README 里那张实测表，口径是接上 OpenClaw 跑公开基准，数字全部公开可查： WideSearch 通过率从 33% 提到 50%，token 反而省了六成 SWE-bench 连续 50 个任务，58.4% 提到 64.2% PersonaMem 长期记忆测试从 48% 提到 76% 思路是两层式的架构：短期记忆把工具调用日志压成符号图，不保留原始日志；长期记忆把碎片对话提炼成人物画像和场景记忆，不做平铺的向量堆积。全本地运行，零外部 API 依赖，可以直接私有化部署。 GitHub：github.com/TencentCloud/T… 提醒一句，这些数字全是官方自己测的，第三方复现出来之前别当成定论。但记忆层是做 agent 绕不开的一环，这张公开的基准表至少给了第三方一个可以复现验证的对象，等复现结果出来再下结论也不迟。

该算账了：Fable 5 的五折补偿窗口周二刚过期，7 月 8 号起全档位按量计费，每百万 token 输入 10 美元、输出 50 美元。我按一次 200 万输出 token 的重度编码会话算了一笔账：Fable 5 要 100 美元，Opus 4.8 是 50 美元，Sonnet 5 只要 20 美元。 性能差距摆在这里：SWE-bench Pro 上 Sonnet 5 是 63.2%，Opus 4.8 是 69.2%，Fable 5 在 80% 以上。问题在于价格差了五倍，分数差了不到二十个点，重度用户每开一次会话都要衡量值不值，跑批量任务的用户，账单涨幅会非常明显。 17 美元月费的 Pro 里还留着两个便宜选项兜底：Sonnet 5 挂着 2 美元/10 美元的优惠价到 8 月底，Opus 4.8 是 5 美元/25 美元。Anthropic 这套排布意图很直白，日常任务用便宜的，难题再上旗舰，这个阶梯明显是按用户的付费意愿设计的。 接下来就看重度用户会不会真的转向便宜档位：你日常跑的任务里真需要旗舰模型的比例能有多少，我猜连两成都不到。

@ColorfulToilet 蹬冒烟它😆😆

@bainianAI 站起来登

三次重置，我直接 sol 极高极快！ 往死里迪拜！

额，这几个意思？我没干啥特殊的啊？