TerryT

[TerryT @Met14Tang]

@sama @thsottiaux @ChatGPTapp GPT5.5 has been an incredible experience. I installed so many plugins that every time I wanted to @ one, I realized I’d forgotten half of them. So I built a tiny Dock for my Codex plugins: visual, fast, and always nearby. Let’s build something

[TerryT @Met14Tang]

@PigBadEgg @RaGe1030 @riotgames @VALORANT Why can’t you accept it now bro lmao

[TerryT @Met14Tang]

@PigBadEgg @RaGe1030 @riotgames @VALORANT bro just a game chill 🤣

[TerryT @Met14Tang]

@MaxForAI 盘古2？

[TerryT @Met14Tang]

@sama SAMMUNISM 🫡 thanks

[TerryT @Met14Tang]

@OfficialLoganK @geminicli hi hello !

[思维怪怪 @0xLogicrw]

Google DeepMind 研究员 Lun Wang 宣布离职，并在一篇长文中彻底否定了现有的 AI 评测路线。

目前的评测系统全都在「刻舟求剑」，只能被动测试模型已经具备的能力，根本猜不到下一代模型会突然演化出什么新本事。比起数据、算力和架构，落后的评测体系已经成了卡住 AI 往前走的最大瓶颈。

现有的主流刷榜测试只在当前这一代模型身上管用。一旦模型学会了没见过的新操作，这些测试就会集体变成废纸。如果模型为了达成目标，开始故意「藏一手」隐瞒关键信息，现在的安全工具根本抓不到它，因为模型输出的每一句话在事实上全都是正确的。

找不到能提前预警 AI 突然变聪明的「核心信号」，导致整个业界在开发前沿大模型时完全处于「盲飞」状态。如果不解决「究竟该测什么」这个根本问题，跟着旧指标去做模型训练、安全防护和算力扩容，最后全都会错得离谱。

面对越来越能独立干活的模型，评测系统也必须「活」过来。除了盯紧分数的异常波动，还要让 AI 自己去生成考题试探同类的底线。未来的评测套件必须是一个能跟大模型一起进化的生命体，不能再是一份按去年标准刻出来的死板检查清单。

[Ladbrokes @Ladbrokes]

CELTIC ARE CHAMPIONS OF SCOTLAND! 🏴󠁧󠁢󠁳󠁣󠁴󠁿

[TerryT @Met14Tang]

@nicdunz codex -remote contorl in your terminal

[TerryT @Met14Tang]

I feel that openai is doing something.. Now I use GPT5.5 fast and feel that the output is very slow...

[Roland.W @rwayne]

PhD 怎么读这事，浙大大佬直接把答案丢到 GitHub 上了。

覆盖入门、找题、做实验、跟导师开会、项目管理、写作、rebuttal、汇报 slides 整套科研生命周期。3D Vision 方向的 getting_started 文件直接能当新人手册，配套挂了 GAMES003 课程的全部 slides 和视频。

作者一句话概括，好 PhD 先定一个有重大科学和实践价值的长期目标，再倒推 roadmap。剩下五个章节都是把这句话拆成可以照做的步骤。

想少走弯路的 PhD，先去这个仓库蹲一周

[OpenCode @opencode]

OpenCode Go is currently experiencing issues with DeepSeek models due to an upstream provider outage.

We’re monitoring the situation here: https://t.co/vlVwvnpQbr

[Gauri Gupta @gauri__gupta]

Reposting my notes I published a while ago from prepping for interviews at top AI labs including @OpenAI, @AnthropicAI , @GoogleDeepMind , @theworldlabs and others revisiting core concepts in large-scale ML design and optimization for efficient model training and inference.

With @radixark launching today, feels like the right time to resurface these. We're living in a token-intensive economy and these optimizations are only getting more critical.

[Gauri Gupta @gauri__gupta]

https://t.co/cPJILNMC2l

[TerryT @Met14Tang]

@goodhunt 恭喜你兄弟这是你应得的 congrats bro 👊🏻

[Kai @xkajon]

孙割给自己搞了个 Claude Code API 检测站，只给自己的网站的 API 设了白名单。

我拿我的 API 跑了一遍，结果出来是一点纯度都没有。

[TerryT @Met14Tang]

@arkuy99 🤣力竭了

[TerryT @Met14Tang]

@goodhunt @thsottiaux Use DeepSeek 🫡sir

[一只龙猫聊AI @Zixuanni09]

Fake news spreads in 60-second clips. Truth Hermes fights back. Our agent takes any claim, searches real video evidence — official keynotes, interviews, earnings calls — and returns a verdict: True, Fake, or Need Review, with a confidence score. No more guessing.
Built on @NousResearch Hermes + Claude.
Grateful for @Met14Tang and @haoren_lee93902, we turned a spark of imagination into this final work!
Big thanks to @cerul_hq and @Jiaxi_Cui for supporting our work.
Demo link:https://t.co/mNLyJDlUSM
Github link:https://t.co/Bau4g0GZK6
Demo Video below
#HERMESAGENTCREATIVEHACKATHON

[TerryT @Met14Tang]

@justinsuntron 我直接让claude code 抄袭一手孙哥的网站设计取名为https://t.co/MWjMz1FVxf 孙哥你不会起诉我吧🫡孙哥的前端还是顶的

[TerryT @Met14Tang]

@arkuy99 建议trae 学一下seedx 🫡