开启时会模糊预览图，关闭后正常显示
I don't want to undermine anyone's work but can someone explain how can a model score > 50% on humanity's last exam and still fail on ALL my λ-calculus questions, including the easiest ones that even gpt-oss gets right?
Kind / Bend / HVM / INets / λCalculus
Perhaps the greatest shortcoming of the present generation of university presidents in the United States is their inside focus on administration, on money-raising, and so on. Yet no other administrator in the large university is free to establish contact with the students who are the university’s “customers.” Alienation of the students from the administration is certainly a major factor in the student unhappiness and unrest - Peter Drucker, The Effective Executive (1966)
Wonderer. Amor fati. Scaling trust.
📣I'll be in Delhi mid-December--to speak on 14th at https://t.co/zAkyByzBmV in @iitdelhi. I may have some time the couple of days after that and am happy to meet with folks to discuss things #AI.. (DM/email).
AI researcher & teacher @SCAI_ASU. Former President of @RealAAAI; Chair of @AAAS Sec T. Here to tweach #AI. YouTube Ch: https://t.co/4beUPOmf6y Bsky: rao2z
我在使用 Claude Code 工作过程中，这几个 MCP 给我帮了很多忙，推荐给小伙伴使用，特别是 Context7 以及 Memory 这两个，他们作用如下，有需要可以去安装配置一趴。 Context7 – 提供最新、版本特定的文档与代码示例，可即时检索任何库或框架的使用方式，适合保持 API 更新同步。 Memory – 跨会话存储与读取上下文信息，让模型具备持久记忆能力，更好地延续项目语境。 GitHub – 直接访问代码仓库、Issue 与 Pull Request，无需切换环境即可查看、修改或引用代码。 Markitdown – 将 PDF、Word、Excel、图片、音频等多种文件格式快速转换为 Markdown，方便阅读与编辑。 Playwright – 用于浏览器自动化测试与数据提取，适合验证前端交互或批量抓取结构化内容。 Serena – 面向 AI 编码代理的语义代码检索与编辑工具，提升模型的代码理解与上下文感知能力。
Father of Pake • MiaoYan • Mole • XRender
smell of fresh cut grass at the park, pokemon red save wiped, zelda game guide book creased at the edges, trampoline burns, dunkaroos frosting, scholastic book fair money in a ziplock, tamagotchi beep at 3am, ice cream truck gumball eyes, blockbuster aisles, vhs rewind sound, capri sun straw stabbing, mtv spring break, machinima, metallic gel pens, heeleys, triple collateral on highrise , mini camcorder, neopets login music, falador park, recon armor, AIM door creak, razor scooter ankle scar, lunchables pizza cold, mcdonalds orange drink, third eye blind on a burned cd, airsoft wars, lava lamp hum, game boy advance sp click, pokemon cards in plastic sleeves, mew under the truck, cheat codes written on binder paper, cereal box prizes, harry potter midnight release, slamball on spike tv, 3D televisions, legends of the hidden temple, through the fire and the flames by dragonforce, warheads tongue burn, scooby-doo fruit snacks, diving board at neighborhood pool, disney channel wand intro, yugioh duel disk, halo 3 energy sword duel, joseph kony, nintendogs barking, easy-bake oven smell, sidewalk chalk, bottom of otterpops, slinky on carpet, old spice spray, livestrong bracelets, birthday sleepovers, club penguin iceberg, now what’s what i call music 5, ipod touch jailbreak, “hags”, naruto, glow-in-the-dark stars on ceiling, cereal marshmallow dust, scooter downhill race, that soft tv pop when it turned off, and sparklers fizzing out far too soon
investing @a16z // curating https://t.co/ssslqn6eo7
When new methods can't match the resources that have been poured into old ones, stagnation ensues. We're there now in AI. The solution is to compare methods at similar stages of development.
Professor of computer science at UW and author of '2040' and 'The Master Algorithm'. Into machine learning, AI, and anything that makes me curious.