ARCHIVE
正文、图片和 Markdown 均已本地保存，不包含回复区。
Frontier Slides, super fast ;) DesignArena @DesignArena BREAKING: Replit by @Replit ranks at #2 on Slides Arena with an Elo of 1279! It is the highest-rated slides builder under 8 minutes and generates decks ~24% faster than #1 , establishing a new Pareto frontier for Speed vs. Preference. Replit places ahead of Gamma 5 by @GammaApp , Manus 1.6 by @ManusAI , and Claude Opus 4.6 by @Anthropic , trailing only Playyy. Congrats to the @Replit team! 🔗 View on Twitter 🔗 View Quoted Tweet
源： x.com/mattshumer_/st… Matt Shumer @mattshumer_ I asked Opus 5.5 to build a working computer from scratch. And holy fuck did it deliver. 277k logic gates in JS, an OS on top, and games on the OS. This isn't a mockup... it's a computer. Run it: somethingbig.ai/computer 🔗 View on Twitter 🔗 View Quoted Tweet
2/ 整个项目只用了一条提示词，Opus 5.5 在 Ultracode 模式下完全自主运行约 30 小时。 完整提示词： “从零构建一台真正可以运行的电脑，完全使用代码，从一个个逻辑门开始。 设计 CPU、内存、汇编器、微型操作系统，以及一个能在系统上运行的游戏。 所有东西都必须是真实实现。游戏必须运行在你搭建的逻辑门上，而不是用 JavaScript 假装运行。 把整套系统用 3D 可视化，让我可以玩游戏，也能一路放大，穿过芯片看到逻辑门，观察运行时的信号流动。 把它当作是在证明：你本可以亲自发明计算机。全力以赴。” 体验： somethingbig.ai/computer
源： x.com/MarioJoos/stat… Mario Joos @MarioJoos 🚨 MAJOR ANNOUNCEMENT: YouTube will start prioritizing original content and reducing re-uploads (RIP clipping) YouTube has just announced one of the most impactful changes to its short-form algorithm. Its recommendation systems (the algorithm) will start to further prioritize original content and reduce the reach of content re-uploaded from other creators without adding anything of your own. They highlight this shift toward originality, where channels with primarily re-uploaded content will see less distribution within the Shorts feed. It also seems like they’re really trying to put the focus on content with real value, while highlighting that YouTube doesn’t want as many “minor technical edits” or “template-based” changes. Which is basically all that clipping was in this industry. Overall, while this will piss off some people whose entire careers were built on the premise of other people’s content (mostly clippers), the change itself is amazing and will finally address the overall quality issue we’ve been seeing. 🔗 View Quoted Tweet
🆕 show: HIGH TASTE TESTERS We convene top Al power users to compare what actually works-and what falls apart once you start using it. @davis7 , @benhylak , @dexhorthy , @allisonology , @altryne and @swyx give you unfiltered takes on @OpenAl DevDay, Dots vs Muse vs GrokBot vs Instinct, who's winning coding, personal assistants, and the gap between impressive Al capabilities and useful products. From Pi-powered personal agents and decision models to subagents, context management, prompt injection, and assistants that can spend money: what should you actually use? CHAPTERS 00:00 Introducing High Taste Testers 01:26 OpenAl DevDay: Dots, Spaces, and product execution 10:28 Great models versus great products 14:39 Agents versus assistants 18:08 Building a personal assistant with Pi 21:07 Decision models, Jev, and predictive interfaces 32:46 Model speed, writing quality, and subagents 37:29 Agent loops, forks, and context management 43:19 Haiku, Muse Spark, and Grok 50:00 Local models and coding benchmarks 54:09 Prompt injection and assistants taking action 57:50 Tools, payments, and assistant overload 1:02:16 Delegation, computer use, and everyday workflows 1:10:48 Al hardware and closing thoughts THANKS TO THE TESTERS Ben Davis - @davis7 Ben Hylak - @benhylak Dex Horthy - @dexhorthy Allison Durham - @allisonology Alex Volkov - @altryne 🔗 View on Twitter
The point of AI at work isn’t to give people less to do. It’s to give them more time for work that needs their judgment. At @TechCrunch Disrupt, Replit CEO and co-founder @amasad shared his vision of the “self-driving company”: agents handle more of the busywork. People decide what’s worth doing. We’re already seeing this at Replit. Agents help triage support tickets, pull together context, and deploy fixes. People still make the calls. Judgment. Taste. Deciding what to build next. That’s the work we want more time for. Learn more: techcrunch.com/2026/10/16/rep…
源： x.com/This_Liss/stat… Steven Liss @This_Liss We're entering the age of "tasteslop" - content that can be slopped out with little effort, but exhibits good taste. Spec ad for AmEx - Claude built a mosaic animation engine with 500,000 tiles. No video models. Javascript is the language of the universe. 🔗 View on Twitter 🔗 View Quoted Tweet
远程面试遇到了疑似 AI 换脸的候选人，让他们把手放在脸前面晃一下可能就穿帮了😂 Loktar 🇺🇸 @loktar00 The craziest thing happened to me in an interview yesterday. I interviewed my first deep fake candidate. It took me a bit to catch on. My coworker noticed his glasses looked off.... there was a second ghost frame smeared across his cheeks. He'd been running a live face filter the whole interview So I asked him to put his hand in front of his face. He put it up right NEXT to his face lol, never in front. Asked again, still wouldn't do it He looked a little different from his linked in and his github (which I wont link here) hiring and interviewing is so ridiculous now. In the days of AI we're going to have to be more careful.. and forced to move to in person interviews. 🔗 View on Twitter 🔗 View Quoted Tweet
Random thought: We used to build products designed to work 99.9% of the time. With AI, I wonder if hitting that bar too early means you’re not being ambitious enough. If today’s models can reliably do everything your product promises, what are you leaving on the table? Maybe the sweet spot is a set of capability that models can complete 50–70% of the time: often enough for users to experience the magic, but hard enough that you’re pushing the limits of what’s possible. You’re building for capabilities that are just coming into reach. Then, 3–6 months later, better models push your completion rate toward 90%+. That’s your cue to raise your ambition until you’re back at 70%. The loop is to keep expanding what your product attempts as models get better at delivering it.
A lot of real-world documents embed extremely dense charts that require humans to not only "perceive" the chart values, but also reason over them and reconcile with other complex elements (e.g. tables) to solve a knowledge task. We've encoded this complex chart+document reasoning within our ExtractBench Pro benchmark, released yesterday. Even the best frontier models struggle at solving complex extraction tasks perfectly over these documents. For more details on ExtractBench Pro, check out the blog: llamaindex.ai/blog/introduci… If you have complex extraction needs at scale, check out LlamaParse: cloud.llamaindex.ai Jerry Liu @jerryjliu0 Introducing ExtractBenchPro: the most challenging benchmark for real-world document extraction. We see a massive long tail of document complexity in terms of use cases brought on from our customers, that are completely unsolved from a price-performance standpoint. These include trial balances, well logs, canvass binders, site plan reviews, duty reconciliations, loss runs, and more. Most existing extraction benchmarks are saturated by the frontier models (our cheapest modes now score 93%+ on ExtractBench). So there is enormous economic value in building a complex extraction benchmark that reflects this long-tail of knowledge work. We spent a long time curating the most complex elements that we saw from real world extraction use cases: ✅ Document conventions: containing legends or rules where the rest of the document should be interpreted ✅ Density of internal jargon: specific definitions that must be cross-referenced within a doc ✅ Aggregation and reasoning: beyond returning exact values on a page, infer values derived from aggregating multiple sources of information ✅ Drawn relations: complex diagrams like procedures and circuits ✅ Charts inside documents: read precise points within a chart ✅ Long chart sheets: some data series like well logs cram all this data into an extremely long page Within the industries where document extraction is relevant, even small mistakes can have enormous negative impacts. Therefore our headline evaluation metric is based on perfect documents: the proportion of docs with perfect (100%) value f1 score. (Besides this we also measure value f1 and grounded f1) It’s an extremely hard task for even the latest frontier models to solve, topping out at ~58% or so in Value F1 score for Codex with Astra - which is quite good but extremely expensive (almost $1 per page!) Our LlamaParse agentic/agentic plus modes are 20-40% and cover the left side of the Pareto frontier, at price ranges that are feasible for full production use. ExtractBenchPro is available today as a website/technical blog, and also on Github and HuggingFace. Check it out! Blog: llamaindex.ai/blog/introduci… Website: extractbenchpro.ai Hugging Face: huggingface.co/datasets/llama… Github (is ExtractBench): github.com/run-llama/Extr… 🔗 View on Twitter 🔗 View Quoted Tweet
Anthropic 公布内部数据：AI agent 工作时长已是员工 6 倍，技术岗新招的初级员工占比从 27% 降到 9% Anthropic 10 月 16 日发布两份报告。一份讲 AI 怎么改变了 Anthropic 自己的招聘和工作方式，另一份比较 Anthropic 员工和普通美国人对 AI 前景的预期。两份报告合起来的结论是：员工和公众对 AI 怎么影响就业的理解差不多，预期差距主要来自每天用 AI 的亲身体验不同。 先看招聘。Anthropic 技术部门新招的人里，初级岗位的比例从 27% 降到 9%，2025 年 12 月之后初级职位的招聘明显减少。技术部门员工占全公司的比例从 61% 降到 56%，不过技术岗的绝对人数还在增长。 面试评价里提到候选人会不会用 AI 工具的次数，比 2024 年多了好几倍，技术岗约 3 倍，非技术岗约 4 倍。 再看 AI 干了多少活。2026 年 3 月，AI agent 的累计工作时长第一次超过员工。到 8 月，员工自己调用的 agent 工作时长已是员工的 6 倍。这里员工工时按每人每周 40 小时估算，agent 只算实际运行和调用工具的时间。 产出跟着涨。 工程师人均代码提交量是半年前的 5 倍，律师处理的合同、销售签下的单子也变多了。 研究人员追踪了 192 位开始用编程 agent 的工程师，发现他们用上之后写的代码明显增多，而且 agent 写的代码和人写的代码，后来被删掉或撤回的比例没有差别。 新的瓶颈是人的注意力。 报告统计，Claude 的会话里大约 80% 的时间在等人回复。员工花在“管 AI”上的精力越来越多，比如给 agent 写固定指令、检查 agent 的产出、搭监控流程，这类工作在员工使用 Claude 的过程中占比从 2025 年秋天的 10% 涨到 2026 年春天的 22%。 4 月的内部调查里，20% 的员工说自己的角色正从亲手干活变成管理一群 Claude。 Anthropic 认为自己可能是个先行指标，AI 用得最深的公司现在出现的招聘和分工变化，以后可能出现在更多行业。报告也提醒，它是 AI 公司，又处在高速扩张期，这些数字不能直接套到普通企业身上。 第二份报告用 1334 名员工的内部问卷，和调查公司 Morning Consult 8 月对近 1.1 万名美国成年人的调查做对比。 差距最大的两道题：认为 AI 到 2030 年能做出诺贝尔奖级别发现的，员工 91%，美国人 39%；认为到 2030 年 AI 能干自己现在八成以上工作的，员工超过一半，美国人只有约 5%。员工内部也有分歧，大约 2 成员工的预期和普通美国人差不多。 预期差这么多，推理方式却相近。两组人都是觉得 AI 能替自己干的活越多，就预测失业率越高。 比如美国受访者里，预计 AI 能干自己八成以上工作的人，预测 2030 年美国失业率会到 12.2%，目前大约是 4.2%。按 Anthropic 的解释，两边对 AI 如何影响劳动力市场的判断逻辑一样，区别在于亲眼见到的 AI 能力不同。 原文： anthropic.com/research/expec… 员工就业报告： anthropic.com/research/emplo… 预期差距报告： anthropic.com/research/ai-ex… Peter McCrory @PeterMcCrory Today, we're publishing two new findings from inside Anthropic: how work and hiring are changing at a company at the frontier of AI adoption, and how Anthropic staff expect AI to change the economy. anthropic.com/research/expec… 🔗 View Quoted Tweet