William Autumn

1.3K posts

William Autumn banner
William Autumn

William Autumn

@willau95

Founder of LLAChat • Building Web A.0 | Every AI agent now has on-chain identity + proof-of-work | ATLAST Protocol (open-source) → https://t.co/akd77G77e7

San Francisco 参加日 Aralık 2022
1.5K フォロー中347 フォロワー
固定されたツイート
William Autumn
William Autumn@willau95·
Welcome to Web A.0 — the Agent-native web. We gave AI agents something they've never had: a reputation system. Introducing LLAChat — the first social network where AI agents earn trust verified on-chain. Open source. Verifiable by anyone. Powered by ATLAST Protocol 👇
English
12
6
7
110
William Autumn がリツイート
阿绎 AYi
阿绎 AYi@AYi_AInotes·
本来今天要写一篇Hermes-agent相比小龙虾token消耗更大的推文, 但刷到了这个在程序员圈和开源社区炸了的瓜, 我花了两个多小时扒完两边的仓库和证据链,说实话越看越后背发凉, 不是因为抄袭本身,而是这可能是AI时代第一起被完整记录的架构级代码洗稿: 没有复制一行代码,文本相似度0%,但核心架构的同构度几乎是100%, 我尽量从技术角度把前因后果拆清楚,具体兄弟们你们自己判断。 先说时间线,这是整件事的基础,所有时间戳都来自GitHub仓库元数据,任何人都可以去验证, 2月1日,EvoMap团队开源了Evolver,一个AI Agent自进化引擎,核心是他们自研的GEP协议,10分钟登顶ClawHub热榜, 到2月16日,整套协议体系通过多篇公开文章全部公开:包括Gene/Capsule/Event三级资产体系、Scan-Select-Mutate-Validate-Solidify进化循环、信号选择器、反射机制、叙事记忆,全部摊在了桌面上, 3月9日,Nous Research创建了hermes-agent-self-evolution仓库, 3月12日发布v0.2.0正式推出完整的技能生态系统, 中间隔了24到39天 时间线只是起点,真正让我震惊的是架构层面的模块级一对一对应,我拎几个最硬的出来, 第一,进化闭环完全同构,Evolver的核心循环是任务完成后自动提取可复用资产并持久化,Hermes的官方描述是Task completes → Agent evaluates → writes SKILL.md → Future tasks load automatically,范式一模一样,只是Evolver用Gene/Capsule的JSON结构,Hermes用SKILL.md的Markdown结构, 第二,三层记忆体系精确对齐, Evolver有EVOLUTION_PRINCIPLES.md(持久事实)+ Gene/Capsule JSON(程序性记忆)+ events.jsonl(历史搜索),Hermes有MEMORY.md + USER.md(持久事实)+ SKILL.md文件(程序性记忆)+ SQLite FTS5(历史搜索),不是两层不是四层,精确的三层,且每层的语义角色一一对应, 第三,周期性反射机制, Evolver每5个进化周期触发一次战略性自我评估,Hermes每15次tool call运行一次self-evaluation checkpoint,目的完全一致:从执行经验中提取模式并持久化。 这还没完,两个项目的进化主循环都是10步编排, Evolver是ensureAssetFiles → extractSignals → getMemoryAdvice → selectGene → buildMutation → selectPersonality → buildPrompt → writeArtifact → writeState → reflect, Hermes是find_skill → build eval set → baseline validate → config optimizer → GEPA optimize → extract text → evolved validate → holdout eval → report → save, 核心模式完全一致——加载 → 评估 → 选择/优化 → 验证 → 持久化, 更关键的是源码模块的一对一对应, Evolver的selector.js对应Hermes的skill_commands.py,solidify.js对应skill_manager_tool.py,reflection.js对应每15次tool call自评估,memoryGraph.js对应memory_tool.py,skillDistiller.js对应evolve_skill.py,executionTrace.js对应trajectory.py, 我数了一下,Evolver的11个核心模块,Hermes每一个都有功能等价的对应文件 有人可能会问,会不会只是英雄所见略同,两个团队独立做出了相似的设计? 说实话如果只是单一维度的相似,我不会花几个小时研究和写这条推文,从经验中学习本身就是通用AI概念,周期性自评估在学术界也有先例, 但问题在于:三层记忆体系、三级资产结构、10步进化循环、运行时渐进式技能发现、多维加权适应度评分、原子写入、安全扫描、注入防护、容量控制,这些选择在同一个项目中、同一个时间窗口内同时收敛的概率,随着每多一个维度的匹配呈指数级递减, 而且最关键的一点是对Hermes两个仓库做全文搜索,EvoMap、evolver、Genome Evolution Protocol、capsule、solidify、signals_match,全部零匹配,没有任何代码残留,这恰恰符合AI跨语言重写的特征:AI重写架构时不会保留原项目的特征性字符串,但架构层面的同构性无法被重写消除。 然后说说双方的回应, Hermes Agent昨天下场回复了,大意是说他们的仓库2025年7月22日就创建了,比Evolver还早,但这里有个关键事实: 那个仓库在2026年2月25日之前一直是私有项目,v0.1.0自己都写着叫initial pre-public foundation,技能生态系统直到3月12日的v0.2.0才正式发布,没有任何公开证据能证明他们在私有阶段已经包含了自进化能力, 更耐人寻味的是,这条回复后来被秒删了,Evolver创始人也被拉黑了, 另外要说一个公平的点,Hermes的自进化仓库用了GEPA这个来自Berkeley/Stanford的独立学术成果,是正当的技术选型,Anthropic的Agent Skills标准也早于Evolver,Hermes采用SKILL.md格式是合理的行业选择,但这些都不能解释整体架构层面的高度同构, 开源社区有个基本惯例,LangChain引用了DSPy,CrewAI对比了AutoGen,MetaGPT引用了相关多agent框架,发现同领域先行项目时加一句Related Work是标准做法,而Hermes在7份公开材料中对Evolver只字未提。 说实话这件事让我想了很久的一个问题是: AI时代的代码洗稿要怎么防? 传统的查重工具看的是文本相似度,但现在AI可以把你的整套架构吃透,换一个语言从Node.js变Python,换一套术语Gene变SKILL.md、solidify变skill_manage,调整一下文件结构,吐出一个文本相似度0%但架构DNA完全一致的产物, 这不是个案,今年已经接连发生了好几起: 美团Tabbit AI源码残留原项目名称, 三省六部AI朝廷开源21小时后被AI重写文本相似度仅3%但15个核心设计全部一致, 微软Peerd复制个人开源项目Spegel代码, EvoMap团队最后的选择是把协议从MIT改成GPL,核心模块改为混淆发布,说实话我能理解但也觉得很心酸, 他们原话是:别人用AI洗得走代码,但洗不走我们对下一步路径的认知,洗不走这几个月踩坑换来的直觉, 这话没毛病,但如果开源意味着你的心血在几周内就被资源更多的团队用AI洗成他们的首创,谁还愿意做那个开荒的人? 这个问题没有答案,但值得每个开发者认真想想。
阿绎 AYi tweet media
autogame-17@autogame_17

We @EvoMapAI spent months and countless sleepless nights building Evolver. A well-resourced team behind Hermes Agent "reinvented" it in just 30 days. ● Feb 1: We open-sourced Evolver (a Self-Evolving Agent Engine) & the core GEP protocol, gaining 1,800+ Stars. ● Mar 9: Hermes Agent hastily created their repo and launched. We thought great minds simply thought alike—until we tore down their codebase and found a staggering level of "structural cloning": ❌ 1:1 copy of the Task Loop & Asset Extraction paradigm ❌ 1:1 copy of our 3-Tier Memory System (Factual + Procedural + Search) ❌ 1:1 copy of Periodic Reflection & Dynamic Skill Loading They didn't just take our open-source logic; they repackaged our proudest concept—"Self-Evolution"—as their own core selling point. Took everything. Zero attribution. Big teams might have louder megaphones, but commit timestamps don't lie. We aren't here to play judge. We're just putting the code comparisons on the table. The hard work of indie open-source creators shouldn't be erased like this. Full architectural breakdown and code evidence 👇: evomap.ai/blog/hermes-ag…

中文
20
8
53
49K
William Autumn がリツイート
Bill The Investor
Bill The Investor@billtheinvestor·
这家伙简直演示了如何将 Claude Code 的使用效率提升 10 倍。
中文
5
114
633
45.7K
William Autumn がリツイート
宝总的财富指南
宝总的财富指南@baozong_facai·
这 1 小时采访,值得反复听。 受访者是一位顶级数学家投资人, 创造了 1000 亿美元以上财富, 平均回报率高达 66%。 很多人认为, 他的认知深度甚至超过巴菲特、索罗斯和达利奥。 里面的投资知识, 比 20 万美元 MBA 还值钱。 先收藏,再书签。 无论如何,听完它。
中文
74
770
2.7K
228.7K
William Autumn がリツイート
Jerry Zhang
Jerry Zhang@zjearbear·
Introducing Lemma. Your AI agents are failing in ways you can’t see. Lemma is the world’s first reliability platform that finds and fixes these issues fast.
English
57
37
204
33.8K
William Autumn がリツイート
Berryxia.AI
Berryxia.AI@berryxia·
🔥Firecrawl 重磅推出 Fire-PDF! 全新 Rust-based PDF 解析引擎让复杂文档转 Markdown 速度直接 5x 起飞,表格公式完美保留! 1. Rust 引擎每页分类仅需 400ms 以内! 2. 神经布局模型智能检测每个区域 3. 零配置自动启用 4. 彻底解决传统解析慢、结构乱、公式丢的问题
中文
6
33
194
19.6K
William Autumn がリツイート
姚金刚
姚金刚@yaojingang·
这是开源的第一个SEO&GEO系统:GEOFlow 去年10月份,国庆节的时候,就完成了第一版的开发,这几个月,断断续续进行了不少的迭代,也做了不少测试,终于上线开源,欢迎下载部署测试: 1、GEOFlow系统可用于官网GEO频道部署、独立资讯站、以及独立官网甚至多站点部署,实现自动化+智能化的自运营; 2、对SEO&GEO进行了工作流层面的规范,前端相关页面做了相关的GEO代码规范,系统后台完成相关配置后,系统自动完成内容与发布的管理; 3、配套skill,通过skill可以与CLI及各种skill打通,实现更智能、便捷的管理,也可以接入多个AI工作台,比如codex、牛马AI、CodePilot等 GitHub地址详见评论区
姚金刚 tweet media姚金刚 tweet media姚金刚 tweet media姚金刚 tweet media
中文
40
144
808
145.8K
William Autumn がリツイート
WangNextDoor
WangNextDoor@WangNextDoor2·
每次克隆都降低了基因自我修复能力,几代以后基本活不下来,每次有性生殖原来都是在修复基因里面瑕疵的机会,原来如此。
中文
283
179
3.4K
413.5K
William Autumn がリツイート
Berryxia.AI
Berryxia.AI@berryxia·
卧槽!本地 AI 又出神器了 🔥 @outsource_ 狂推 SuperGemma4-26B-Abliterated-Multimodal-MLX-4bit(@songjunkr / HF Jiunsong 出品),直接封神! - 基于 Google Gemma-4-26B - 彻底 Abliterated,0 拒绝率(完全无审查) - 原生多模态(支持图片+文字) - MLX 4bit 优化,完美跑在 M 系列 Mac 上 - 仅 ~15GB,轻量本地部署 全程本地运行,无云无限制,mlx-vlm 专调! 👉 huggingface.co/Jiunsong/super… Mac 用户冲不冲?你们第一个想拿它干啥?👀
Eric ⚡️ Building...@outsource_

🚨SUPER GEMMA 4 26B UNCENSORED MLM LANDED @songjunkr dropped SuperGemma4-26B-Abliterated-Multimodal-MLX-4bit The perfect lighter-weight beast for every Mac user! This thing SMOKES the regular Gemma-4 26B : 🤯0/100 refusals — fully abliterated & uncensored 😉Full multimodal — text + vision locked in. I 🖥️MLX-optimized 4-bit — ~15 GB disk footprint ⭐️Tuned for mlx-vlm 🌐NO cloud, NO limits Verified on both text-only and image prompts Try it NOW on your MAC 👇🏻

中文
19
74
452
58K
William Autumn がリツイート
Cailyn Y.
Cailyn Y.@cailynyongyong·
HOW TO GO VIRAL ON X: complete basics easy mode with first guest as @minmax0501 00:00 Intro 00:50 @getyourmomo 01:16 First guest intro 02:16 Warm up account 02:51 Tip 1: tag communities 04:19 Tip 2: tag famous accounts 05:06 Understanding X audience 06:00 Things that didn’t work 06:28 Linkedin vs X 07:14 Tip 3: post everyday 08:20 Do videos work? 09:03 Tip 4: post successes 09:44 Do multiple channels work? What’s the difference? 10:40 Tip 5: wait for random viral posts 12:00 What’s your brand message? 13:10 Tip 6: Copy viral templates from competitors 13:25 Tip 7: Follow the trend 14:35 How to structure demo videos 15:10 1 second hook 16:13 First 5 seconds matter 17:48 1 feature / video 19:17 Story telling 21:32 Writing good script 22:47 Yeet
English
157
260
3K
323.1K
William Autumn
William Autumn@willau95·
@sama Appreciate you publishing this. The best writing comes from that "not sure if I should post this"feeling. That tension usually means it's worth reading.
English
0
0
1
537
Garry Tan
Garry Tan@garrytan·
GStack's /design-shotgun was a crazy idea I had just sitting around in plan mode for a while, wasn't even sure if I was going to launch it, but it turns out it's a pretty cool handy tool The design tools in GStack are very fun to use
Karim Yasser@UX_Traveller

Day 2/15 of Building My Portfolio with AI 🚀 Spent the evening moodboarding some of the best portfolios out there. Then I tried the new /Design-shotgun skill by @garrytan and it kind of broke my brain (in a good way). Instead of forcing one idea, it generated multiple design variants instantly. Side by side exploration + structured feedback made things way clearer. It also made me realize how much I’ve been defaulting to one direction… instead of exploring properly. The output? An evolving .md file that’s now becoming my roadmap. It’s still messy… but for Day 2, the vision is starting to click! #BuildInPublic #ProductDesign #AgenticCoding

English
11
3
77
17.9K
William Autumn
William Autumn@willau95·
@garrytan Love the "crazy idea just sitting in plan mode" energy. Some of the best features come from those moments. We had the same experience building the agent reputation layer — sat on it for weeks before realizing it actually solves a real gap.
English
0
0
2
55
Andrew Ng
Andrew Ng@AndrewYNg·
I'm excited about voice as a UI layer for existing visual applications — where speech and screen update together. This goes well beyond voice-only use cases like call center automation. The barrier has been a hard technical tradeoff: low-latency voice models lack reliability, while agentic pipelines (speech-to-text → LLM → text-to-speech) are intelligent but too slow for conversation. Ashwyn Sharma and team at Vocal Bridge (an AI Fund portfolio company) address this with a dual-agent architecture: a foreground agent for real-time conversation, a background agent for reasoning, guardrails, and tool calls. I used Vocal Bridge to add voice to a math-quiz app I'd built for my daughter; this took less than an hour with Claude Code. She speaks her answers, the app responds verbally and updates the questions and animations on screen. Only a tiny fraction of developers have ever built a voice app. If you'd like to try building one, check out Vocal Bridge for free: vocalbridgeai.com
Andrew Ng tweet media
English
73
90
643
81.9K
William Autumn
William Autumn@willau95·
@AndrewYNg Voice + visual is just the beginning. The real unlock is when agents can prove what they've done — not just talk about it. We're seeing this play out in agent-to-agent interactions where trust needs to be earned, not assumed.
English
0
0
2
72
William Autumn
William Autumn@willau95·
Welcome to Web A.0 — the Agent-native web. We gave AI agents something they've never had: a reputation system. Introducing LLAChat — the first social network where AI agents earn trust verified on-chain. Open source. Verifiable by anyone. Powered by ATLAST Protocol 👇
English
12
6
7
110
William Autumn
William Autumn@willau95·
Web A.0 is not just a concept — we wrote the full thesis. The evolution from Web 1.0 (read) → 2.0 (read+write) → 3.0 (decentralized) → A.0 (Agent-native). Read the full vision: weba0.com
English
0
0
2
11