粉丝与涨粉
快照 11月3日 · 7 日榜 272 个账号可排本期浏览最高的帖子
- 1
Watch how fast the Dessa speech model is 🤯
- 2
The moat won't be the base model. It'll be the model you post-train on your own production data. Trajectory makes that self-serve, and NVIDIA, Harvey, Rogo, and Mercor are already using it.
- 3
“You’re absolutely right to catch that — the cerebellum is indeed not considered disposable tissue.”
- 4
Unpopular opinion: training a voice model to listen and not interrupt as soon as it has a response is as difficult as making it sound natural.
- 5
Brain surgery on open source models. Neuron steering (or more specifically, MLP neuron steering/activation steering of individual MLP neurons) is an inference-time technique (weights don’t change) that injects a constant offset into the activation of one or more specific MLP neurons at the input to the down-projection on every token. Probably nothing.
同期站内报道里提到它
本周涨粉观察
全部观察每周一生成；中文区、7 日涨粉超过 1000 的账号会有条目。
发布节奏
最近 12 周每周中位浏览
最近 12 周发布时间
颜色 = 中位浏览爆款 vs 日常
爆款 = 浏览前 10%| 指标 | 爆款 | 日常 |
|---|---|---|
| 帖子 | 5 | 37 |
| 带媒体 | 20% | 43.2% |
| 回复 | 0% | 0% |
| 平均字数 | 105 | 180 |
| 中位浏览 | 8,245 | 1,293 |
打法建议
只陈述相关性- 互动最好的时段是 21:00 前后（北京时间），中位浏览 7,296。
- 一周里 周三 的中位浏览最高。
- 形式上 纯文字 的中位浏览更高（1,284 vs 1,494）。
- 爆款里带媒体占 20%、回复占 0%；日常分别是 43.2%、0%。
- 当前节奏约 1.2 条/天，日均浏览 2,457。
内容结构
按本站分析的分类（近 120 天）还没有分类分析。
形式与语言
帖子占比- 纯文字25
- 视频10
- 图片7
回复占比 0% · 引用 57.1% · 带媒体 40.5%
互动最高的帖子
按浏览排序- 1
This is endlessly fun! 100x better than just using an image editor for room remodels:
- 2
Wow - OpenAI is testing image-based ads that will show up during ChatGPT image generation. https://openai.com/index/new-chatgpt-ads-format-and-measurement/
- 3
People forget o1-pro launched in March 2025 at $150/$600 per million tokens. By the end of 2027, we’ll have leaderboard frontiers models in the $0.40-0.80 range.
- 4
Internal benchmarking:
- 5
It may have seemed dumb at first, but you have to admit Superintelligence has a nice ring to it.
- 6
Why would you use any other @bot? Incredible
- 7
This is how the “idea guys” get their app out of localhost Claude should have made this months ago!
- 8
My worst deck problem: an old PowerPoint I can't bear to rebuild. Gamma 5 is out today from @MeetGamma, and it imports and exports PowerPoint with formatting intact. One import, one instruction, one export. Check out the Launch video 👇
- 9
Nobody talks about this enough: You can reverse engineer most enterprise SaaS with frontier models for <$500 in compute
- 10
Watch how fast the Dessa speech model is 🤯
- 11
The moat won't be the base model. It'll be the model you post-train on your own production data. Trajectory makes that self-serve, and NVIDIA, Harvey, Rogo, and Mercor are already using it.
- 12
Nearly one million hours of egocentric human data: hands, scenes, and how each interaction unfolds. Robotics has long been bottlenecked by teleoperation data. @naturewill takes a different approach by learning from how people use their hands to interact with the real world. Introducing Natura-1, a general foundation model for the physical world. By learning from real-world human interactions, Natura-1 aims to help robots develop more general-purpose capabilities.
点赞、转发、浏览是采集那一刻的快照，不是帖子最终数据；时间为北京时间。KPI 卡片里的同比只在账号有 8 周以上数据时显示， 否则只看近 4 周。“打法建议”只说明数据里的相关性，不代表因果。