与「JK」相关的搜索结果

我去，在 GitHub 上直接拿下了 13K 颗星！ 有人搞了个 AI 新闻雷达出来，能帮你从整个互联网里挖出特别有价值的信息！ 1、把 Telegram、X、GitHub、RSS、Reddit 以及 Hacker News 等一堆信息源全聚到了一起。 2、AI 会自动筛选、打分、去重，把那些真正值得一看的内容给你留下来。 3、每天的中英文简报它自己就能生成，还会顺手把事件背景和网友讨论给你补上。 🔗： ⁠

你敢让 agent 在你那个装满钱包插件的浏览器里随便点吗？ 我不敢 那就换个思路，给它一台自己的电脑，模型也关在本地跑 CopilotKit OpenBot 每个 bot 自带一台电脑，浏览器 文件 工具都是它自己的 Mac 和 Windows 有安装包，不碰终端也能装，就是得登个账号 ashhart TensorFold Mac 上 brew 一行装好，不用配 Python 跑起来就是一个本地接口，OpenAI 和 Anthropic 两种格式都认 我想先拿小模型给项目公告做摘要，跑不动再换大的 QwenLM qwen-code 终端里写代码的 agent，快三万星了 模型随便换，Ollama 本地的也行 还能挂到 Telegram 和飞书上，躺床上也能派活 活儿让 agent 在它那台电脑上干，签名这总得自己来

20 free dune dashboards memecoin traders actually use what's inside: > paid 15 callers $5,054,233. since oct 12 the wallets that bought their calls are down $899,112 > $1,000 in the median call is $515 an hour later and $263 a day later > 45.4% of called coins peak within 5 minutes. 40.9% were called after their top callers 1.) → what paid each caller vs what their followers made 2.) → median price 1 minute to 1 day after a call, by caller 3.) → how many minutes a called coin has before it peaks 4.) → daily revenue since jan 2024, launchpad vs pumpswap 5.) → PUMP buybacks and fees, 58 charts 6.) → $535,052,472 since sep 2025. 86.6% of last month's came from coins launched the same day 7.) → minutes until the dev's first sell, month by month 8.) → wallets ranked by coins launched and how many graduated find wallets 9.) → profitable wallets across 31 trading bots 10.) → the last 24h top runners and the wallets that bought them early 11.) → profitable wallets on coins created in the last month 12.) → filter wallets by win rate, SOL earned and median hold time 13.) → filter by median ROI, win rate and scalp ratio the market 14.) → the solana ecosystem in 58 charts 15.) → memecoin volume by chain. solana does 51.9% of it 16.) → which trading app does the volume on each chain 17.) → DEX volume by chain and by DEX. 11,975 stars perps, predictions, airdrops 18.) → hyperliquid fees and revenue, 61 charts 19.) → polymarket vs kalshi: volume, users, fees 20.) → every big airdrop and who claimed it, 489 charts dead or broken. stop bookmarking these: > adam_tehc/memecoin-wars (1,935 stars), axiom, gmgn, fomo, the-robinhood-trenches → last day of data sep 22 > whale_hunter/dex-trading-bot-wars (2,107 stars) → 24 of 37 charts failed > uwusanauwu/perps → 15 of 21 charts failed how to check any dune dashboard in 5 seconds: > the line under the title says "updated 2d ago" and how many charts failed > a yellow "last day of data" line on top = dead > look at the last date on the first chart bookmark this

🚨 BREAKING: ACTIVE SHOOTER with a semi-automatic rifle reported on the East Side of Midtown in NYC MULTIPLE casualties reported, including NYPD officers Suspect is thought to be wearing a bulletproof vest Pray for our boys in blue! 🙏🏻

Former White House press secretary Karoline Leavitt will join Fox News as an on-air contributor next month, weeks after stepping down from her prior role to spend more time focusing on her family.

又来安利「Hands-On Large Language Models」 作者 @JayAlammar @MaartenGr 全书绘制了近 300 张自制图，对应理解概念和原理、架构非常有帮助，作者们叫它 "The Illustrated LLM Book"。 Alammar 的图解系列从 2018 年起就是很多人理解 Transformer、BERT、GPT 的入口，被斯坦福、MIT 等课程引用；这本书把那套「先给直觉和图，再给代码」的讲法系统化沉淀成了 12 章内容。 12 章可以分成三段，这个分法也呼应了 @Nils_Reimers 在书评里的概括，生成式、表示式、检索式三类应用： 理解模型（1–3 章）：语言模型是什么、token 和嵌入怎么工作、Transformer 内部长什么样。第 3 章就是 Illustrated Transformer 的更新版。 表示与检索（4、5、8、10、11 章）：文本分类、聚类与主题建模、语义搜索和 RAG、自己训练嵌入模型、微调 BERT 类表示模型。这是本书区别于其他 LLM 书的部分，大多数书聚焦生成，这条嵌入线来自 Grootendorst 做 BERTopic 的一手经验，讲的是「模型不生成文本、而是把文本变成向量」的那一半应用。 生成与微调（6、7、9、12 章）：提示工程、高级生成技术（采样、结构化输出等）、多模态、微调生成模型。 Alammar 对第 3 章的图解： 代码开源在这： 两位作者还参加了一次 @DeepLearningAI 课程：