Herrington Darkholme

17.9K posts

Herrington Darkholme banner
Herrington Darkholme

Herrington Darkholme

@hd_nvim

🌐 Frontend Vimmer ⚒️Open Source with @typescript @vuejs and @rustlang 🚀 https://t.co/SLjF6No9qD is my hobby project

Katılım Temmuz 2011
2.7K Takip Edilen7.4K Takipçiler
Sabitlenmiş Tweet
Herrington Darkholme
Herrington Darkholme@hd_nvim·
Announcing the book "𝑀𝑎𝑠𝑡𝑒𝑟𝑖𝑛𝑔 𝑎𝑠𝑡-𝑔𝑟𝑒𝑝"!🦉 The book is a structured deep-dive: we start with the mental models of ASTs, move to rule composition, and end with advanced usage. It’s the place where design principles and philosophy are explained. Link below ↓
Herrington Darkholme tweet media
English
6
6
53
6.2K
Herrington Darkholme
Herrington Darkholme@hd_nvim·
It's huge for open source * OSS model company gets paid * Commercial company gets fine tuned model as their most * Serving company gets deals Virtuous cycle
Kimi.ai@Kimi_Moonshot

Congrats to the @cursor_ai team on the launch of Composer 2! We are proud to see Kimi-k2.5 provide the foundation. Seeing our model integrated effectively through Cursor's continued pretraining & high-compute RL training is the open model ecosystem we love to support. Note: Cursor accesses Kimi-k2.5 via @FireworksAI_HQ ' hosted RL and inference platform as part of an authorized commercial partnership.

English
0
0
10
828
投资TALK君
投资TALK君@TJ_Research·
另外再补充一点，从美联储的视角里，市场已经在帮他们紧缩了，他啥都不用干，金融环境指数就紧缩了，有时候他们会FOLLOW THROUGH，顺着市场走，有时候他们会纠偏。所以年底加不加息不是重点，重点时市场现在定价了加息，就已经达到了加息的作用。今天就到这里了，周末愉快。
投资TALK君@TJ_Research

2021年11月，联储会议，鲍威尔：也许资产价格下跌可以让人们回到劳动力市场，解决劳动力短缺的问题，参与度底下的问题。后来回顾时，鲍威尔说当时看到ECI环比增速破新高，觉得一定要加息了 22年年初，市场认为美联储最多只能加2次，忘了吗？多少人说那么高的债务加多了经济要崩，结果加了几次？ 现在，市场定价今年50%可能加息一次，我认为债市是错的，美联储通常都是LOOK THROUGH油价的上涨。即使油价在高位，高油价对于需求是致命的打击，特别是服务通胀。当然一切的一切，最重要是劳动力市场，现在和22年完全相反，风险是向下的，不是22年向上的，涨工资更是不可能的事情。BUY $IEI

中文
17
1
149
26.9K
Tesla
Tesla@Tesla·
We – design the chips & hardware – make the cars w/ said hardware – collect real-world data at scale – train the real-world AI model – built (& continue to expand) the massive supercomputer cluster that trains it – deploy AI directly to millions of robots on wheels All that is shared with @Tesla_Optimus for broader applications in both the physical & digital world
English
631
1.8K
12.3K
869.1K
Lee Robinson
Lee Robinson@leerob·
Yep, Composer 2 started from an open-source base! We will do full pretraining in the future. Only ~1/4 of the compute spent on the final model came from the base, the rest is from our training. This is why evals are very different. And yes, we are following the license through our inference partner terms.
Fynn@fynnso

was messing with the OpenAI base URL in Cursor and caught this accounts/anysphere/models/kimi-k2p5-rl-0317-s515-fast so composer 2 is just Kimi K2.5 with RL at least rename the model ID

English
336
166
2.5K
1.1M
Herrington Darkholme
Herrington Darkholme@hd_nvim·
fear index is still 17 :/ needs more spaces for a single digit
English
0
0
0
307
Labelray
Labelray@Labelray·
其实今天梦到和我妈吵架了，但是也没有伤心，好像已经接受了我妈不会在现实生活中出现的这个事实了
中文
2
0
9
660
小马驹 哒哒哒 哒哒哒
小马驹 哒哒哒 哒哒哒@jjgfscbd·
想骗中登上床，一般怎么表达比较好？ 光靠G奶已经无法诱惑到这老男人了…
中文
162
3
286
170.9K
Frank Wang 玉伯
Frank Wang 玉伯@lifesinger·
CLI 这波热点，是流量逻辑。因为 for agents 的述事成了硅谷的吹捧，同时确实有 agents 想通过 CLI 来调用工具，于是传统工具们纷纷穿上新衣服，希望被调用，来获取流量或资本关注。 然而 CLI 化有很难解决的问题： 1. 参数爆炸问题。GUI 的功能远胜 CLI，要压缩成 CLI 参数，只能疯狂做减法。能敢于做减法的，已经是有认知的老板。没认知的老板，只会强压团队：所有功能都必须 CLI 化。可悲的是，用户根本不这么想。比如用户会在 IM 里按照自己对传统工具的 GUI 用法来描述需求，结果 agent 会发现：CLI 只能骗用户，或调调最基础的功能，稍微复杂的，还得是 computer use. 可用户已经急躁得骂娘了。 2. 更严重的问题上，CLI 化很难跑通商业模式。此处太敏感，省略一万字。 更好的解法，是别安装本地软件了，一切云端 API 化。Agent 也很擅长调用 API 的，而且很考验 API 的接口易用性和商业模式设计。越易用的 API，能有越多模型能调用成功。商业模式越好的 API，则越能更长久为用户提供性美价廉甚至免费的服务。 提供 API 最有经验的公司，就是 SaaS 公司。优秀 SaaS 公司最核心的资产，也是 API。并不是界面。 Agent 时代， 需要 CLI，然而更需要的，是优秀的 API 型 SaaS 服务。
中文
33
11
143
28K
Cyandev
Cyandev@unixzii·
MCP：tool call 套壳 Skill：prompt 套壳（prompt 宏） CLI：freeform 参数的 tool Tool Call：真正的模型能力
中文
18
22
276
27.9K
GeekPlux
GeekPlux@geekplux·
3 月 28 号在上海，有来面基的吗
GeekPlux tweet media
中文
10
0
26
2.7K
꥟ ENTELECHEIA ꥟
꥟ ENTELECHEIA ꥟@cyrnosofia·
Hay una persona que usa él seudónimo de "Marisa Kirisame" de Touhou y está haciendo un doctorado en ciencias del cómputo en la universidad de Utah, lo más gracioso es que está haciendo aportes a Machine Learning
꥟ ENTELECHEIA ꥟ tweet media꥟ ENTELECHEIA ꥟ tweet media꥟ ENTELECHEIA ꥟ tweet media꥟ ENTELECHEIA ꥟ tweet media
Español
41
580
4.7K
122.4K
Herrington Darkholme
Herrington Darkholme@hd_nvim·
Lecun's last FAIR paper
alphaXiv@askalphaxiv

Yann LeCun and his team dropped yet another paper! "V-JEPA 2.1: Unlocking Dense Features in Video Self-Supervised Learning" In this V-JEPA upgrade, they showed that if you make a video model predict every patch, not just the masked ones AND at multiple layers, they are able to turn vague scene understanding into dense + temporal stable features that actually understands "what is where". This key insight drove improvements in segmentation, depth, anticipation, and even robot planning.

Français
0
0
0
361
Herrington Darkholme
Herrington Darkholme@hd_nvim·
#ast_grep_trick You can use the `PreToolCall` hook in your coding agent to check bash command, by not just blocking but providing context on every `Write|Edit|Bash` Credits to xero in ast-grep discord
Herrington Darkholme tweet media
English
0
0
3
378

Keşfet

@flowtype @MetaOpenSource @rustlang @TJ_Research @Tesla @Tesla_Optimus @leerob @Labelray