<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>xv01</title>
    <link>https://qua.name/xv01/</link>
    <description></description>
    <pubDate>Mon, 01 Jun 2026 19:43:56 +0200</pubDate>
    <item>
      <title>My Wish 遗嘱</title>
      <link>https://qua.name/xv01/my-wish</link>
      <description>&lt;![CDATA[如果我死在 x 年 , zx31415 死在 y 年 , 2026 &lt; x &lt; y , 现在 2026 年 , zx31415 不知道我的银行卡密码 , 要如何在我死后 , 让 zx31415 可以处理我银行卡的钱 ( 捐给数学研究/教育机构或者其他合法用途 , 我完全信任 zx31415 )&#xA;&#xA;...... AI 回答 ......&#xA;&#xA;!--more--&#xA;&#xA;设立遗嘱（推荐方式）:&#xA;    操作：在您生前，前往公证处订立一份公证遗嘱。在遗嘱中明确写明：“本人名下所有银行账户（可列出银行名称和账号）内的资金，全部捐赠给[具体数学研究/教育机构名称，如中国数学会]，用于支持数学研究与教育事业。”&#xA;    执行：您去世后，zx31415作为知情人，可以持您的死亡证明和这份公证遗嘱，联系您指定的捐赠机构。该机构可以凭借遗嘱向银行申请办理资金划转手续。整个过程无需zx31415知道您的银行卡密码。&#xA;    优点：手续相对简单，成本较低，是大众最常用的方式。&#xA;&#xA;设立生前信托:&#xA;    操作：在您生前，与信托公司签订信托合同，将您的银行账户资金作为信托财产。合同中约定：“委托人（您）去世后，受托人（信托公司）应将信托财产全额捐赠给[具体数学研究/教育机构名称]。”&#xA;    执行：您去世后，信托公司会根据合同自动执行捐赠，zx31415无需介入，也无需知道密码。&#xA;    优点：执行更高效，能完全规避遗产继承的繁琐程序，隐私性更强。&#xA;&#xA;核心要点：&#xA;必须生前完成：无论是遗嘱还是信托，都必须在您意识清醒时完成法律手续。&#xA;指定具体机构：遗嘱或信托中必须写明具体的、合法的捐赠机构全称，避免模糊不清。&#xA;zx31415的角色：zx31415的作用是“通知”或“联系”，而非“操作”。他/她只需在得知您去世后，将您的遗嘱或信托信息告知捐赠机构即可。&#xA;&#xA;------&#xA;&#xA;我想帮助自身天赋不高但喜欢数学的人学数学 ( 尤其是早年教育被耽误想弥补的那些人 ) 最好是在我生前自己设立一个机构做这件事 &amp; 我死后银行卡里的钱直接转给这个机构作为以我名字命名的基金 ?&#xA;&#xA;...... AI 回答 ......&#xA;&#xA;可以。&#xA;&#xA;您完全可以在生前设立一个以您名字命名的基金会或专项基金，专门帮助天赋不高但热爱数学、希望弥补早年教育的青少年，您去世后，银行卡里的钱直接转入该机构作为基金。&#xA;&#xA;操作路径如下：&#xA;&#xA;设立基金会或专项基金：&#xA;    独立基金会：您作为发起人，向民政部门申请设立一个“非公募基金会”，最低原始基金为200万元人民币。基金会的宗旨和业务范围明确为“资助和培养对数学有热情但基础薄弱的青少年”。您可以在基金会章程中规定，该机构以您的名字命名。&#xA;    专项基金：更简单的方式是，您向一个已有的、信誉良好的基金会（如中国青少年发展基金会或中国儿童少年基金会）捐赠一笔资金，设立一个“[您的姓名] 数学教育专项基金”。这种方式无需您承担繁琐的运营工作。&#xA;&#xA;资金传承：&#xA;    在您生前，将启动资金注入该基金会或专项基金。&#xA;    在您的遗嘱或信托合同中明确写明：“本人名下所有银行账户的全部余额，在本人去世后，由执行人/受托人一次性转入[基金会全称]，作为‘[您的姓名] 数学教育基金’的永久性本金，用于资助符合条件的青少年数学教育项目。”&#xA;&#xA;这样，您的机构在您生前就能开始运作，帮助青少年，而您去世后，资金会自动注入，确保您的慈善事业永久延续。&#xA;&#xA;------&#xA;&#xA;我银行卡只有 10 万左右 , 只能用信托的形式 , 不够设立基金 ?&#xA;&#xA;...... AI 回答 ......&#xA;&#xA;是的，您银行卡里的10万元不足以独立设立一个具有法人资格的基金会，因为根据《基金会管理条例》，非公募基金会的最低注册资金为200万元人民币。&#xA;&#xA;但是，您完全可以用这10万元实现您的慈善目标，最合适的途径是设立专项基金。&#xA;&#xA;操作方式：&#xA;选择母基金会：寻找一家您信任的、已注册的大型基金会（如中国青少年发展基金会、中国儿童少年基金会、地方性慈善总会等）。&#xA;设立专项基金：与该基金会协商，用您的10万元在其名下设立一个“[您的姓名] 数学教育专项基金”。&#xA;签订协议：双方签订协议，明确该基金的宗旨为“资助热爱数学但基础薄弱、希望弥补的青少年”，并规定基金的命名、使用规则和监督机制。&#xA;资金传承：在您的遗嘱或信托中写明，您去世后，银行卡的全部余额将捐赠给该基金会，用于扩充“[您的姓名] 数学教育专项基金”的本金。&#xA;&#xA;这样，您的名字和慈善理念就能通过这个专项基金永久传承，专门帮助您想帮助的青少年群体。&#xA;]]&gt;</description>
      <content:encoded><![CDATA[<p>如果我死在 x 年 , zx31415 死在 y 年 , 2026 &lt; x &lt; y , 现在 2026 年 , zx31415 不知道我的银行卡密码 , 要如何在我死后 , 让 zx31415 可以处理我银行卡的钱 ( 捐给数学研究/教育机构或者其他合法用途 , 我完全信任 zx31415 )</p>

<p>...... AI 回答 ......</p>


<ol><li><p><strong>设立遗嘱（推荐方式）</strong>:</p>
<ul><li><strong>操作</strong>：在您生前，前往公证处订立一份<strong>公证遗嘱</strong>。在遗嘱中明确写明：“本人名下所有银行账户（可列出银行名称和账号）内的资金，全部捐赠给[具体数学研究/教育机构名称，如中国数学会]，用于支持数学研究与教育事业。”</li>
<li><strong>执行</strong>：您去世后，zx31415作为知情人，可以持您的死亡证明和这份公证遗嘱，联系您指定的捐赠机构。该机构可以凭借遗嘱向银行申请办理资金划转手续。整个过程<strong>无需zx31415知道您的银行卡密码</strong>。</li>
<li><strong>优点</strong>：手续相对简单，成本较低，是大众最常用的方式。</li></ul></li>

<li><p><strong>设立生前信托</strong>:</p>
<ul><li><strong>操作</strong>：在您生前，与信托公司签订信托合同，将您的银行账户资金作为信托财产。合同中约定：“委托人（您）去世后，受托人（信托公司）应将信托财产全额捐赠给[具体数学研究/教育机构名称]。”</li>
<li><strong>执行</strong>：您去世后，信托公司会根据合同自动执行捐赠，zx31415无需介入，也无需知道密码。</li>
<li><strong>优点</strong>：执行更高效，能完全规避遗产继承的繁琐程序，隐私性更强。</li></ul></li></ol>

<p><strong>核心要点</strong>：
*   <strong>必须生前完成</strong>：无论是遗嘱还是信托，都必须在您意识清醒时完成法律手续。
*   <strong>指定具体机构</strong>：遗嘱或信托中必须写明<strong>具体的、合法的</strong>捐赠机构全称，避免模糊不清。
*   <strong>zx31415的角色</strong>：zx31415的作用是“通知”或“联系”，而非“操作”。他/她只需在得知您去世后，将您的遗嘱或信托信息告知捐赠机构即可。</p>

<hr>

<p>我想帮助自身天赋不高但喜欢数学的人学数学 ( 尤其是早年教育被耽误想弥补的那些人 ) 最好是在我生前自己设立一个机构做这件事 &amp; 我死后银行卡里的钱直接转给这个机构作为以我名字命名的基金 ?</p>

<p>...... AI 回答 ......</p>

<p>可以。</p>

<p>您完全可以在生前设立一个以您名字命名的基金会或专项基金，专门帮助天赋不高但热爱数学、希望弥补早年教育的青少年，您去世后，银行卡里的钱直接转入该机构作为基金。</p>

<p><strong>操作路径如下</strong>：</p>
<ol><li><p>设立基金会或专项基金：</p>
<ul><li><strong>独立基金会</strong>：您作为发起人，向民政部门申请设立一个“非公募基金会”，最低原始基金为200万元人民币。基金会的宗旨和业务范围明确为“资助和培养对数学有热情但基础薄弱的青少年”。您可以在基金会章程中规定，该机构以您的名字命名。</li>
<li><strong>专项基金</strong>：更简单的方式是，您向一个已有的、信誉良好的基金会（如<strong>中国青少年发展基金会</strong>或<strong>中国儿童少年基金会</strong>）捐赠一笔资金，设立一个“<strong>[您的姓名] 数学教育专项基金</strong>”。这种方式无需您承担繁琐的运营工作。</li></ul></li>

<li><p><strong>资金传承</strong>：</p>
<ul><li>在您生前，将启动资金注入该基金会或专项基金。</li>
<li>在您的<strong>遗嘱或信托合同</strong>中明确写明：“本人名下所有银行账户的全部余额，在本人去世后，由执行人/受托人一次性转入[基金会全称]，作为‘[您的姓名] 数学教育基金’的永久性本金，用于资助符合条件的青少年数学教育项目。”</li></ul></li></ol>

<p>这样，您的机构在您生前就能开始运作，帮助青少年，而您去世后，资金会自动注入，确保您的慈善事业永久延续。</p>

<hr>

<p>我银行卡只有 10 万左右 , 只能用信托的形式 , 不够设立基金 ?</p>

<p>...... AI 回答 ......</p>

<p>是的，您银行卡里的10万元不足以<strong>独立设立</strong>一个具有法人资格的基金会，因为根据《基金会管理条例》，非公募基金会的最低注册资金为200万元人民币。</p>

<p><strong>但是，您完全可以用这10万元实现您的慈善目标</strong>，最合适的途径是<strong>设立专项基金</strong>。</p>

<p><strong>操作方式</strong>：
1.  <strong>选择母基金会</strong>：寻找一家您信任的、已注册的大型基金会（如中国青少年发展基金会、中国儿童少年基金会、地方性慈善总会等）。
2.  <strong>设立专项基金</strong>：与该基金会协商，用您的10万元在其名下设立一个“<strong>[您的姓名] 数学教育专项基金</strong>”。
3.  <strong>签订协议</strong>：双方签订协议，明确该基金的宗旨为“资助热爱数学但基础薄弱、希望弥补的青少年”，并规定基金的命名、使用规则和监督机制。
4.  <strong>资金传承</strong>：在您的遗嘱或信托中写明，您去世后，银行卡的全部余额将捐赠给该基金会，用于扩充“[您的姓名] 数学教育专项基金”的本金。</p>

<p>这样，您的名字和慈善理念就能通过这个专项基金永久传承，专门帮助您想帮助的青少年群体。</p>
]]></content:encoded>
      <guid>https://qua.name/xv01/my-wish</guid>
      <pubDate>Tue, 24 Mar 2026 12:15:32 +0100</pubDate>
    </item>
    <item>
      <title>Use blog, not social media</title>
      <link>https://qua.name/xv01/use-blog-not-social-media</link>
      <description>&lt;![CDATA[Some aged posts:&#xA;全世界都有信息越来越少, 质量越来越低, 越来越难以被检索到的情况,&#xA;大概 英文的 下降速度 略慢 而已&#xA;以后知道去什么小众社区找到需要的东西, 可能会成为一个有用技能, 这不新鲜, 2005年以前就是这样的 ...&#xA;01-17-2022&#xA;&#xA;!--more--&#xA;&#xA;irfanview, the strongest graphics software in history, has been developed since 1996 and is still updated to this day&#xA;running on all versions of Windows&#xA;only 3.56MB &lt;- nothing wrong with decimal point position or unit&#xA;free for individuals, $12 for businesses&#xA;only one person wrote it&#xA;indeed a legend&#xA;https://x.com/virushuo/status/1454880197694697474&#xA;&#xA;a large number of basic projects of the Internet and software&#xA;is silently maintained by Nordics&#xA;there is no money and no name&#xA;prob caused by: people have a good attitude in Nordic life&#xA;if they don’t do it someday, it’s definitely fun&#xA;https://x.com/virushuo/status/1461385272357265415&#xA;&#xA;我唯一的希望就是以后Linus可千万别抱怨内核维护者越来越少&#xA;毕竟从他的回信里面一些客客气气对这个问题表示异议的人仅仅是因为拼音名字就被称作troll和付费演员&#xA;几年前我写过文章说开源会永远存在, 并且说大部分基金会注册在美国并不影响开源发展, 今天我知道我错了&#xA;https://x.com/virushuo/status/1849277775514915235&#xA;&#xA;[privacy, security]&#xA;Clubhouse 的房间录音满天飞了, 这倒是拿事实证明了 telegram 的看法,&#xA;end-to-end 只有两人聊天 且 固定单设备 才有意义, 多人场景加密 和 各种限制 只是自欺欺人&#xA;https://x.com/virushuo/status/1358463550793142282&#xA;&#xA;PS: RMS and Phone Use&#xA;Richard Stallman, often referred to as RMS, is known for his strong stance on privacy and security, particularly in the context of free software and avoiding proprietary technology. He advocates for using computers with fully free software and avoids using mobile phones due to their potential as surveillance devices. Instead, he prefers to use a Thinkpad x200 computer with a free initialization program (Libreboot) and a free operating system (Trisquel GNU/Linux)&#xA;Stallman emphasizes the importance of security and privacy, recommending practices like using 2FA (two-factor authentication) for online accounts, although he has been critical of 2FA&#39;s reliance on email or phone providers, which he views as less secure and out of the user&#39;s control&#xA;He is also known for his careful approach to connecting to the internet, avoiding portals that require identification and preventing nonfree JavaScript code from running on his devices&#xA;&#xA;------&#xA;&#xA;全世界都有信息越来越少, 质量越来越低, 越来越难以被检索到的情况,&#xA;大概 英文的 下降速度 略慢 而已&#xA;以后知道去什么小众社区找到需要的东西, 可能会成为一个有用技能, 这不新鲜, 2005年以前就是这样的 ...&#xA;01-17-2022&#xA;https://x.com/virushuo/status/1483290406322417666&#xA;&#xA;最近发现中文几大新闻网站都多了个新毛病, 用手机浏览器看会把大部分内容折叠起来, 要下app才能看, 过去点那个箭头就会打开折叠, 现在悄悄改了不让点了,&#xA;我看得不多, 但偶尔需要查个东西还挺烦,&#xA;解决方案: 用 firefox android noscripts 插件干掉所有 js,&#xA;开放互联网死于 app&#xA;https://x.com/virushuo/status/1343630981870350338&#xA;&#xA;BBS可能是互联网上最重要的发明, 过去是, 今天仍然是,&#xA;很遗憾中文世界的 BBS 所剩无几 ( 别跟我说天涯什么的 ) 但台湾的 ptt 和 香港的 连登 算是保住了中文 bbs 最后的阵地,&#xA;如果你不明白我说什么, 有个纪录片推荐看看：&#xA;BBS the Documentary [Full HD] &#xA;The BBS (bulletin board system) scene of the 80s and 90s was a magical time. Long before the Internet escaped from the lab, connected the planet and redefine...&#xA;08-24-2019&#xA;https://x.com/virushuo/status/1165159809844486145&#xA;&#xA;人总是得经过一些什么才会学到 &#34;谣言是信息不完全状态的补充&#34; 以及 &#34;机构必须谨慎, 但个人行为可以提前进行&#34;,&#xA;但是这个经验几乎无法传递, 只能看着年轻人从辟谣嘲笑恐慌再经历一次&#xA;01-22-2020&#xA;https://x.com/virushuo/status/1220134072032342016&#xA;&#xA;我之前说过, 海盗湾和 sci-hub 才是真正在互联网上被追杀的服务 ;&#xA;不仅没人提供服务, 创始人还只能在有限地区活动, 出国可能就被抓了引渡美国, 政治法律商业全部用上对付他们, 结果怎么样? 人家还活着 ;&#xA;gab 和 parler 啥也不算, 用不着给他们哭丧, 真关心互联网开放性的还是得关心前面这俩 ;&#xA;https://x.com/virushuo/status/1349812261045547014&#xA;&#xA;[Wikipedia, Openstreetmap] 都是在那个特殊年代人们才有的梦想, 并且实现了 ;&#xA;如果今天说想做一个完全没广告, 众人共创, 全球共用多语言, 并且没有盈利模式全靠捐款的东西, 说出来根本没人相信了, 比说明年移民火星听起来都不可信, 至少得加上会发个币听起来才真实点 ...&#xA;2021-12-02&#xA;https://x.com/virushuo/status/1466532949919109123&#xA;&#xA;&amp; delusion of &#34;democracy&#34; in internet&#xA;  从自由软件运动到维基、从暗网再到最近的零网，都被捆绑上了强烈的技术乌托邦设定。创始人和早期用户们往往是教育良好、熟悉代码的极客，相信网络超越于现实，后者的问题可以利用前者解决。正如媒体运动研究者Todd Wolfson所言，这些人与线下日常政治的脱节，影响到了他们对技术前景的看法。往往正是对技术平台民主潜质的乐观主义，让用户们忘记了践行日常生活的革命。最终，美好的期待一次次落空，新技术一个接一个被体制化。&#xA;  更何况，早期万维网那种繁荣的民主，往往是中产白人青年内部契约的产物，他们当年并没有准备好迎接多元社会的挑战。1993年，整个互联网上只有不到1000个站点，没有政治极化、没有过滤泡泡，每个新加入的独立站点，都自然可以吸引全网的目光。到了1996年，77%的美国成年人也还没有连上互联网。早期理想主义者的成功，建立在发展不均、资源稀缺、精英垄断的基础之上，与真正并肩战斗的国际主义相去甚远。&#xA;]]&gt;</description>
      <content:encoded><![CDATA[<p>Some aged posts:
全世界都有信息越来越少, 质量越来越低, 越来越难以被检索到的情况,
大概 英文的 下降速度 略慢 而已
以后知道去什么小众社区找到需要的东西, 可能会成为一个有用技能, 这不新鲜, 2005年以前就是这样的 ...
01-17-2022</p>



<p>irfanview, the strongest graphics software in history, has been developed since 1996 and is still updated to this day
running on all versions of Windows
only 3.56MB &lt;– nothing wrong with decimal point position or unit
free for individuals, $12 for businesses
only one person wrote it
indeed a legend
<a href="https://x.com/virushuo/status/1454880197694697474" rel="nofollow">https://x.com/virushuo/status/1454880197694697474</a></p>

<p>a large number of basic projects of the Internet and software
is silently maintained by Nordics
there is no money and no name
prob caused by: people have a good attitude in Nordic life
if they don’t do it someday, it’s definitely fun
<a href="https://x.com/virushuo/status/1461385272357265415" rel="nofollow">https://x.com/virushuo/status/1461385272357265415</a></p>

<p>我唯一的希望就是以后Linus可千万别抱怨内核维护者越来越少
毕竟从他的回信里面一些客客气气对这个问题表示异议的人仅仅是因为拼音名字就被称作troll和付费演员
几年前我写过文章说开源会永远存在, 并且说大部分基金会注册在美国并不影响开源发展, 今天我知道我错了
<a href="https://x.com/virushuo/status/1849277775514915235" rel="nofollow">https://x.com/virushuo/status/1849277775514915235</a></p>

<p>[privacy, security]
Clubhouse 的房间录音满天飞了, 这倒是拿事实证明了 telegram 的看法,
end-to-end 只有两人聊天 且 固定单设备 才有意义, 多人场景加密 和 各种限制 只是自欺欺人
<a href="https://x.com/virushuo/status/1358463550793142282" rel="nofollow">https://x.com/virushuo/status/1358463550793142282</a></p>

<p>PS: RMS and Phone Use
Richard Stallman, often referred to as RMS, is known for his strong stance on privacy and security, particularly in the context of free software and avoiding proprietary technology. He advocates for using computers with fully free software and avoids using mobile phones due to their potential as surveillance devices. Instead, he prefers to use a Thinkpad x200 computer with a free initialization program (Libreboot) and a free operating system (Trisquel GNU/Linux)
Stallman emphasizes the importance of security and privacy, recommending practices like using 2FA (two-factor authentication) for online accounts, although he has been critical of 2FA&#39;s reliance on email or phone providers, which he views as less secure and out of the user&#39;s control
He is also known for his careful approach to connecting to the internet, avoiding portals that require identification and preventing nonfree JavaScript code from running on his devices</p>

<hr>

<p>全世界都有信息越来越少, 质量越来越低, 越来越难以被检索到的情况,
大概 英文的 下降速度 略慢 而已
以后知道去什么小众社区找到需要的东西, 可能会成为一个有用技能, 这不新鲜, 2005年以前就是这样的 ...
01-17-2022
<a href="https://x.com/virushuo/status/1483290406322417666" rel="nofollow">https://x.com/virushuo/status/1483290406322417666</a></p>

<p>最近发现中文几大新闻网站都多了个新毛病, 用手机浏览器看会把大部分内容折叠起来, 要下app才能看, 过去点那个箭头就会打开折叠, 现在悄悄改了不让点了,
我看得不多, 但偶尔需要查个东西还挺烦,
解决方案: 用 firefox android noscripts 插件干掉所有 js,
开放互联网死于 app
<a href="https://x.com/virushuo/status/1343630981870350338" rel="nofollow">https://x.com/virushuo/status/1343630981870350338</a></p>

<p>BBS可能是互联网上最重要的发明, 过去是, 今天仍然是,
很遗憾中文世界的 BBS 所剩无几 ( 别跟我说天涯什么的 ) 但台湾的 ptt 和 香港的 连登 算是保住了中文 bbs 最后的阵地,
如果你不明白我说什么, 有个纪录片推荐看看：
BBS the Documentary [Full HD]
The BBS (bulletin board system) scene of the 80s and 90s was a magical time. Long before the Internet escaped from the lab, connected the planet and redefine...
08-24-2019
<a href="https://x.com/virushuo/status/1165159809844486145" rel="nofollow">https://x.com/virushuo/status/1165159809844486145</a></p>

<p>人总是得经过一些什么才会学到 “谣言是信息不完全状态的补充” 以及 “机构必须谨慎, 但个人行为可以提前进行”,
但是这个经验几乎无法传递, 只能看着年轻人从辟谣嘲笑恐慌再经历一次
01-22-2020
<a href="https://x.com/virushuo/status/1220134072032342016" rel="nofollow">https://x.com/virushuo/status/1220134072032342016</a></p>

<p>我之前说过, 海盗湾和 sci-hub 才是真正在互联网上被追杀的服务 ;
不仅没人提供服务, 创始人还只能在有限地区活动, 出国可能就被抓了引渡美国, 政治法律商业全部用上对付他们, 结果怎么样? 人家还活着 ;
gab 和 parler 啥也不算, 用不着给他们哭丧, 真关心互联网开放性的还是得关心前面这俩 ;
<a href="https://x.com/virushuo/status/1349812261045547014" rel="nofollow">https://x.com/virushuo/status/1349812261045547014</a></p>

<p>[Wikipedia, Openstreetmap] 都是在那个特殊年代人们才有的梦想, 并且实现了 ;
如果今天说想做一个完全没广告, 众人共创, 全球共用多语言, 并且没有盈利模式全靠捐款的东西, 说出来根本没人相信了, 比说明年移民火星听起来都不可信, 至少得加上会发个币听起来才真实点 ...
2021-12-02
<a href="https://x.com/virushuo/status/1466532949919109123" rel="nofollow">https://x.com/virushuo/status/1466532949919109123</a></p>

<p>&amp; <a href="https://gravitysworm.com/post/163872148252/%E5%8E%9F%E5%A7%8B%E7%9A%84%E7%BD%91%E7%BB%9C%E4%BA%92%E8%81%94%E7%BD%91%E7%9A%84%E6%97%A9%E6%9C%9F%E6%83%B3%E8%B1%A1" rel="nofollow">delusion of “democracy” in internet</a>
&gt; 从自由软件运动到维基、从暗网再到最近的零网，都被捆绑上了强烈的技术乌托邦设定。创始人和早期用户们往往是教育良好、熟悉代码的极客，相信网络超越于现实，后者的问题可以利用前者解决。正如媒体运动研究者Todd Wolfson所言，这些人与线下日常政治的脱节，影响到了他们对技术前景的看法。往往正是对技术平台民主潜质的乐观主义，让用户们忘记了践行日常生活的革命。最终，美好的期待一次次落空，新技术一个接一个被体制化。
&gt; 更何况，早期万维网那种繁荣的民主，往往是中产白人青年内部契约的产物，他们当年并没有准备好迎接多元社会的挑战。1993年，整个互联网上只有不到1000个站点，没有政治极化、没有过滤泡泡，每个新加入的独立站点，都自然可以吸引全网的目光。到了1996年，77%的美国成年人也还没有连上互联网。早期理想主义者的成功，建立在发展不均、资源稀缺、精英垄断的基础之上，与真正并肩战斗的国际主义相去甚远。</p>
]]></content:encoded>
      <guid>https://qua.name/xv01/use-blog-not-social-media</guid>
      <pubDate>Tue, 24 Mar 2026 11:07:01 +0100</pubDate>
    </item>
    <item>
      <title>Modernism and Involution</title>
      <link>https://qua.name/xv01/modernism-and-involution</link>
      <description>&lt;![CDATA[马克思 : 宗教 是 人民 的 鸦片&#xA;https://mp.weixin.qq.com/s/x-WItklm4Or1BUV7hTrrRw &#xA;​&#xA;​「有一个例外 是 宗教保守派」  ​https://www.zhihu.com/question/625329652/answer/3261438575&#xA;&#xA;!--more--&#xA;&#xA;「STEM」== 现代宗教&#xA;​&#xA;//   ​维系一个共同的迷信是很难的，不是你打个响指就能办到的。&#xA;//   钻石成功洗脑了全世界的人，bitcoin 能成功让全世界那么多人信，这就是实实在在的本事。中本聪行，你不行，这就叫稀缺。&#xA;https://www.zhihu.com/question/58206379/answer/409991875&#xA;&#xA;「自然可能拒绝被轻易解读，但祂不会主动欺骗人类」&#xA;​&#xA;​  如果 数学公理 被完全信任了 , 能进行怎样的 欺骗 &#xA;https://zhuanlan.zhihu.com/p/22389755&#xA;&#xA;//   程序员应该想象力再丰富一点，不要把目光局限于「系统给了我什么权限」，而是要扩展到「如果我被完全信任了，我能进行什么样的欺骗」。​&#xA;https://jhuo.ca/post/wallinfectiontrustcheating &#xA;&#xA;编译器的 blind spot​ &#xA;&#xA;https://manishearth.github.io/blog/2016/12/02/reflections-on-rusting-trust/&#xA;&#xA;https://research.swtch.com/nih&#xA;&#xA;https://ncatlab.org/nlab/show/The+Blind+Spot]]&gt;</description>
      <content:encoded><![CDATA[<p>马克思 : 宗教 是 人民 的 鸦_片
<a href="https://mp.weixin.qq.com/s/x-WItklm4Or1BUV7hTrrRw" rel="nofollow">https://mp.weixin.qq.com/s/x-WItklm4Or1BUV7hTrrRw</a>
​
​「有一个例外 是 宗教保守派」  ​<a href="https://www.zhihu.com/question/625329652/answer/3261438575" rel="nofollow">https://www.zhihu.com/question/625329652/answer/3261438575</a></p>



<p>「STEM」== 现代宗教
​
// &gt; ​维系一个共同的迷信是很难的，不是你打个响指就能办到的。
// &gt; 钻石成功洗脑了全世界的人，bitcoin 能成功让全世界那么多人信，这就是实实在在的本事。中本聪行，你不行，这就叫稀缺。
<a href="https://www.zhihu.com/question/58206379/answer/409991875" rel="nofollow">https://www.zhihu.com/question/58206379/answer/409991875</a></p>

<p><a href="https://zx31415.wordpress.com/2014/12/10/%E7%88%B1%E5%9B%A0%E6%96%AF%E5%9D%A6%E7%9A%84%E4%B8%8A%E5%B8%9D%EF%BC%8C%E5%88%98%E6%85%88%E6%AC%A3%E7%9A%84%E4%B8%8A%E5%B8%9D/" rel="nofollow">「自然可能拒绝被轻易解读，但祂不会主动欺骗人类」</a>
​
​ ** 如果 数学公理 被完全信任了 , 能进行怎样的 欺骗 **
<a href="https://zhuanlan.zhihu.com/p/22389755" rel="nofollow">https://zhuanlan.zhihu.com/p/22389755</a></p>

<p>// &gt; 程序员应该想象力再丰富一点，不要把目光局限于「系统给了我什么权限」，而是要扩展到「如果我被完全信任了，我能进行什么样的欺骗」。​
<a href="https://jhuo.ca/post/wall_infection_trust_cheating" rel="nofollow">https://jhuo.ca/post/wall_infection_trust_cheating</a></p>

<p>编译器的 blind spot​</p>

<p><a href="https://manishearth.github.io/blog/2016/12/02/reflections-on-rusting-trust/" rel="nofollow">https://manishearth.github.io/blog/2016/12/02/reflections-on-rusting-trust/</a></p>

<p><a href="https://research.swtch.com/nih" rel="nofollow">https://research.swtch.com/nih</a></p>

<p><a href="https://ncatlab.org/nlab/show/The+Blind+Spot" rel="nofollow">https://ncatlab.org/nlab/show/The+Blind+Spot</a></p>
]]></content:encoded>
      <guid>https://qua.name/xv01/modernism-and-involution</guid>
      <pubDate>Tue, 24 Mar 2026 10:59:09 +0100</pubDate>
    </item>
    <item>
      <title>LLM - Lean</title>
      <link>https://qua.name/xv01/llm-lean</link>
      <description>&lt;![CDATA[https://mathlib-initiative.org/ from https://www.renaissancephilanthropy.org/funds &#xA;用于「科研」, 商业化 ( 「刚需」 ) 其实是很少的 ... right ?&#xA;&#xA;Ironically interesting to see LLM vibe proving !!&#xA;https://axiommath.ai/territory/from-seeing-why-to-checking-everything&#xA;&#xA;!--more--&#xA;&#xA;For humans, we have decent intuitions. We know problems with heavy case analysis are tedious to enumerate. We know problems requiring one clever construction can make one stuck forever or shout aha! But for machines, we&#39;re still in the dark. A real theory of machine difficulty—what structural features make problems easy or hard for automated provers – seems like a genuinely interesting research direction.&#xA;&#xA;Right now the workflow is simple: humans pose problems, machines find proofs. But you can imagine something more interactive. Machines sanity-checking human ideas in real time. Humans reviewing machine feedback and steering search toward promising directions. Each side complementing the other.&#xA;&#xA;Axiom is building a world where human intuitions are grounded by machine verification and machine verification in turn inspires human intuitions.&#xA;&#xA;这也是一种「涌现」么&#xA;&#xA;还记得 傅立叶变换 实现 同余加法 么 ?&#xA;&#xA;https://x.com/xleaps/status/1627873094814531584&#xA;&#xA;&amp; https://twitter.com/xleaps/status/1627844991824297984&#xA;&#xA;​&#xA;​组合数学的特点在于杂。组合题类型丰富 , 没有「标准工具」且很大程度上依赖非常巧妙的观察 , 因此在数学竞赛和信息学竞赛中也常常被认为是衡量一个人「思维能力」的标准之一&#xA;&#xA;而没有 标准工具 证明灵活 即使不考虑 AI , 仅对编写 Lean 代码的人 , 也会是不小的挑战&#xA;&#xA;​... AlphaProof ​把大量的时间浪费在了细枝末节上 , 而非组合问题中那一两步关键的充满灵感的数学构造 &#xA;https://www.zhihu.com/question/662586636/answer/3577050463&#xA;​&#xA;​组合题相对灵活 , 不需要太多复杂的定理和公式 , 需要的是将问题抽象化和转换的能力 , 这也是为什么 IMO 和 IOI 双栖选手往往在组合题中得分较高的原因 ]]&gt;</description>
      <content:encoded><![CDATA[<p><a href="https://mathlib-initiative.org/" rel="nofollow">https://mathlib-initiative.org/</a> from <a href="https://www.renaissancephilanthropy.org/funds" rel="nofollow">https://www.renaissancephilanthropy.org/funds</a>
用于「科研」, 商业化 ( 「刚需」 ) 其实是很少的 ... right ?</p>

<p>======</p>

<p>Ironically interesting to see LLM vibe proving !!
<a href="https://axiommath.ai/territory/from-seeing-why-to-checking-everything" rel="nofollow">https://axiommath.ai/territory/from-seeing-why-to-checking-everything</a></p>



<p>For humans, we have decent intuitions. We know problems with heavy case analysis are tedious to enumerate. We know problems requiring one clever construction can make one stuck forever or shout aha! But for machines, we&#39;re still in the dark. A real theory of machine difficulty—what structural features make problems easy or hard for automated provers – seems like a genuinely interesting research direction.</p>

<p>Right now the workflow is simple: humans pose problems, machines find proofs. But you can imagine something more interactive. Machines sanity-checking human ideas in real time. Humans reviewing machine feedback and steering search toward promising directions. Each side complementing the other.</p>

<p>Axiom is building a world where human intuitions are grounded by machine verification and machine verification in turn inspires human intuitions.</p>

<p>这也是一种「涌现」么</p>

<p>还记得 傅立叶变换 实现 同余加法 么 ?</p>

<p><a href="https://x.com/xleaps/status/1627873094814531584" rel="nofollow">https://x.com/xleaps/status/1627873094814531584</a></p>

<p>&amp; <a href="https://twitter.com/xleaps/status/1627844991824297984" rel="nofollow">https://twitter.com/xleaps/status/1627844991824297984</a></p>

<p>======
​
​组合数学的特点在于杂。组合题类型丰富 , 没有「标准工具」且很大程度上依赖非常巧妙的观察 , 因此在数学竞赛和信息学竞赛中也常常被认为是衡量一个人「思维能力」的标准之一</p>

<p>而没有 标准工具 证明灵活 即使不考虑 AI , 仅对编写 Lean 代码的人 , 也会是不小的挑战</p>

<p>​... AlphaProof ​把大量的时间浪费在了细枝末节上 , 而非组合问题中那一两步关键的充满灵感的数学构造
<a href="https://www.zhihu.com/question/662586636/answer/3577050463" rel="nofollow">https://www.zhihu.com/question/662586636/answer/3577050463</a>
​
​组合题相对灵活 , 不需要太多复杂的定理和公式 , 需要的是将问题抽象化和转换的能力 , 这也是为什么 IMO 和 IOI 双栖选手往往在组合题中得分较高的原因</p>
]]></content:encoded>
      <guid>https://qua.name/xv01/llm-lean</guid>
      <pubDate>Tue, 03 Feb 2026 12:27:42 +0100</pubDate>
    </item>
    <item>
      <title>LLM code accuracy</title>
      <link>https://qua.name/xv01/records</link>
      <description>&lt;![CDATA[https://www.coze.cn/s/C9xkXQ5_YBI/&#xA;&#xA;!--more--&#xA;&#xA;// 1. https://www.nist.gov/publications/exploring-prompt-patterns-effective-vulnerability-repair-real-world-code-large-language&#xA;​&#xA;while LLMs have inherent limitations in handling complex vulnerabilities independently, they can become effective tools for automated vulnerability repair when guided by carefully crafted prompts&#xA;&#xA;// 2. https://arxiv.org/abs/2503.15341&#xA;​&#xA;​We propose two confidence-based uncertainty measures: Entropy-based and Probability Differential-based methods. When uncertainty is high, UnCert-CoT activates CoT-decoding to generate multiple reasoning paths and selects the final code that exhibits the highest likelihood of correctness.&#xA;&#xA;In contrast, LLM directly generates the code when uncertainty is low.&#xA;&#xA;This uncertainty judgment mechanism allows LLMs to prioritize complex tasks and avoid unnecessary steps in simpler cases, thereby improving overall efficiency and accuracy in code generation.&#xA;]]&gt;</description>
      <content:encoded><![CDATA[<p><a href="https://www.coze.cn/s/C9xkXQ5_YBI/" rel="nofollow">https://www.coze.cn/s/C9xkXQ5_YBI/</a></p>



<p>// 1. <a href="https://www.nist.gov/publications/exploring-prompt-patterns-effective-vulnerability-repair-real-world-code-large-language" rel="nofollow">https://www.nist.gov/publications/exploring-prompt-patterns-effective-vulnerability-repair-real-world-code-large-language</a>
​
while LLMs have inherent limitations in handling complex vulnerabilities independently, they can become effective tools for automated vulnerability repair when guided by carefully crafted prompts</p>

<p>// 2. <a href="https://arxiv.org/abs/2503.15341" rel="nofollow">https://arxiv.org/abs/2503.15341</a>
​
​We propose two confidence-based uncertainty measures: Entropy-based and Probability Differential-based methods. When uncertainty is high, UnCert-CoT activates CoT-decoding to generate multiple reasoning paths and selects the final code that exhibits the highest likelihood of correctness.</p>

<p>In contrast, LLM directly generates the code when uncertainty is low.</p>

<p>This uncertainty judgment mechanism allows LLMs to prioritize complex tasks and avoid unnecessary steps in simpler cases, thereby improving overall efficiency and accuracy in code generation.</p>
]]></content:encoded>
      <guid>https://qua.name/xv01/records</guid>
      <pubDate>Tue, 03 Feb 2026 11:16:57 +0100</pubDate>
    </item>
    <item>
      <title>adventofcode</title>
      <link>https://qua.name/xv01/adventofcode</link>
      <description>&lt;![CDATA[https://adventofcode.com/2022/day/7&#xA;&#xA;https://www.reddit.com/r/adventofcode/comments/zesk40/2022day7_solutions/]]&gt;</description>
      <content:encoded><![CDATA[<p><a href="https://adventofcode.com/2022/day/7" rel="nofollow">https://adventofcode.com/2022/day/7</a></p>

<p><a href="https://www.reddit.com/r/adventofcode/comments/zesk40/2022_day_7_solutions/" rel="nofollow">https://www.reddit.com/r/adventofcode/comments/zesk40/2022_day_7_solutions/</a></p>
]]></content:encoded>
      <guid>https://qua.name/xv01/adventofcode</guid>
      <pubDate>Wed, 19 Nov 2025 03:06:40 +0100</pubDate>
    </item>
    <item>
      <title>Control DIY</title>
      <link>https://qua.name/xv01/control-diy</link>
      <description>&lt;![CDATA[Inverse Kinematics - simple https://www.youtube.com/watch?v=WAsMAeKDc4U&#xA;&#xA;Inverse Kinematics &amp; PID https://pmc.ncbi.nlm.nih.gov/articles/PMC7887770/&#xA;&#xA;Inverted Pendulum - PID - Lego, RaspberryPi, Python https://www.youtube.com/watch?v=WObG2LoSEwQ&#xA;&#xA;Reinforcement learning control of an inverted pendulum using Python, Simulink, and LW-RCP. https://www.youtube.com/watch?v=lM6rYjM6HBU]]&gt;</description>
      <content:encoded><![CDATA[<p>Inverse Kinematics – simple <a href="https://www.youtube.com/watch?v=WAsMAeKDc4U" rel="nofollow">https://www.youtube.com/watch?v=WAsMAeKDc4U</a></p>

<p>Inverse Kinematics &amp; PID <a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC7887770/" rel="nofollow">https://pmc.ncbi.nlm.nih.gov/articles/PMC7887770/</a></p>

<p>Inverted Pendulum – PID – Lego, RaspberryPi, Python <a href="https://www.youtube.com/watch?v=WObG2LoSEwQ" rel="nofollow">https://www.youtube.com/watch?v=WObG2LoSEwQ</a></p>

<p>Reinforcement learning control of an inverted pendulum using Python, Simulink, and LW-RCP. <a href="https://www.youtube.com/watch?v=lM6rYjM6HBU" rel="nofollow">https://www.youtube.com/watch?v=lM6rYjM6HBU</a></p>
]]></content:encoded>
      <guid>https://qua.name/xv01/control-diy</guid>
      <pubDate>Tue, 04 Nov 2025 12:28:46 +0100</pubDate>
    </item>
    <item>
      <title>pytorch batch size exp 2</title>
      <link>https://qua.name/xv01/pytorch-batch-size-exp-2</link>
      <description>&lt;![CDATA[pytorch batch size exp 2&#xA;&#xA;!--more--&#xA;&#xA;// 2023-08&#xA;&#xA;torch&#xA;batch size&#xA;set to 16&#xA;&#xA;slow 一次迭代 = 1.4 seconds ;&#xA;15 or 17 一次迭代 只需要 0.35 seconds ;&#xA;cuda 占用都是 99% ;&#xA;显存都只用了一半 ;&#xA;worker_num = 4 ;&#xA;&#xA;batch size = 8 一次迭代要 0.5 seconds 比设为 15 or 17 慢&#xA;&#xA;一说 : batch size 设成 2 的幂 , 方便显卡分配&#xA;但是现在事实说明 , 不设成 2 的幂反倒更快&#xA;......&#xA;&#xA;盲猜缓存问题，2 的整数次幂可能导致某些情况下缓存命中率下降&#xA;一个 kernel 内部会复用当前 batch 的数据&#xA;&#xA;shared memory bank conflict&#xA;可能是某个 kernel 有 bank conflict&#xA;&#xA;应该是，查了一下，还是 cuda 的经典问题&#xA;&#xA;我记得有一种填充的方法，每行填充一个空格，就可以错开&#xA;&#xA;------&#xA;&#xA;CPU 的 cache 有类似的现象，读取 stride 是 cache 大小的倍数，导致一直出现缓存冲突&#xA;CSAPP 第六章，存储器层次结构&#xA;&#xA;------&#xA;&#xA;以前学信竞的时候 被教导过 数组 不要开 2 的次幂&#xA;最好开 质数 ?&#xA;&#xA;然而 用 数组 造 循环队列 就是要 开 2 的幂 的长度&#xA;才能把 取模运算 优化成 位运算 2%n 可以被优化为 2&amp;(n-1)&#xA;从而提高速度&#xA;&#xA;开 2^n+1 &#xA;然后最后一个用来当临时变量 ?&#xA;]]&gt;</description>
      <content:encoded><![CDATA[<p>pytorch batch size exp 2</p>



<p>// 2023-08</p>

<p>torch
batch size
set to 16</p>

<p>slow 一次迭代 = 1.4 seconds ;
15 or 17 一次迭代 只需要 0.35 seconds ;
cuda 占用都是 99% ;
显存都只用了一半 ;
worker_num = 4 ;</p>

<p>batch size = 8 一次迭代要 0.5 seconds 比设为 15 or 17 慢</p>

<p>一说 : batch size 设成 2 的幂 , 方便显卡分配
但是现在事实说明 , 不设成 2 的幂反倒更快
......</p>

<p>盲猜缓存问题，2 的整数次幂可能导致某些情况下缓存命中率下降
一个 kernel 内部会复用当前 batch 的数据</p>

<p>shared memory bank conflict
可能是某个 kernel 有 bank conflict</p>

<p>应该是，查了一下，还是 cuda 的经典问题</p>

<p>我记得有一种填充的方法，每行填充一个空格，就可以错开</p>

<hr>

<p>CPU 的 cache 有类似的现象，读取 stride 是 cache 大小的倍数，导致一直出现缓存冲突
CSAPP 第六章，存储器层次结构</p>

<hr>

<p>以前学信竞的时候 被教导过 数组 不要开 2 的次幂
最好开 质数 ?</p>

<p>然而 用 数组 造 循环队列 就是要 开 2 的幂 的长度
才能把 取模运算 优化成 位运算 2%n 可以被优化为 2&amp;(n-1)
从而提高速度</p>

<p>开 2^n+1
然后最后一个用来当临时变量 ?</p>
]]></content:encoded>
      <guid>https://qua.name/xv01/pytorch-batch-size-exp-2</guid>
      <pubDate>Wed, 22 Oct 2025 06:25:58 +0200</pubDate>
    </item>
    <item>
      <title>Mul DP</title>
      <link>https://qua.name/xv01/mul-dp</link>
      <description>&lt;![CDATA[&#xA;&#xA;mul.png&#xA;&#xA;Find the maximum product when placing multiplication operators in a number string.&#xA;&#xA;Problem Analysis:&#xA;&#xA;Given:&#xA;A string of N digits (6 ≤ N ≤ 40)&#xA;Need to place K multiplication operators (1 ≤ K ≤ 6)&#xA;Find the maximum product possible&#xA;&#xA;For example, with string &#34;1231&#34; and K=2:&#xA;We need to split it into 3 parts (K+1 parts)&#xA;Find the arrangement that gives maximum product&#xA;&#xA;!--more--&#xA;&#xA;C++ Solution:&#xA;&#xA;include iostream&#xA;include string&#xA;include vector&#xA;include algorithm&#xA;include climits&#xA;using namespace std;&#xA;&#xA;long long getNumber(const string&amp; s, int start, int end) {&#xA;    long long num = 0;&#xA;    for (int i = start; i &lt;= end; i++) {&#xA;        num = num  10 + (s[i] - &#39;0&#39;);&#xA;    }&#xA;    return num;&#xA;}&#xA;&#xA;long long solve(const string&amp; s, int n, int k) {&#xA;    // dpi = maximum product using first i digits with j multiplications&#xA;    vectorvector&lt;long long  dp(n + 1, vectorlong long(k + 2, 0));&#xA;    &#xA;    // Initialize: no multiplication, just the number itself&#xA;    for (int i = 1; i &lt;= n; i++) {&#xA;        dpi = getNumber(s, 0, i - 1);&#xA;    }&#xA;    &#xA;    // Fill the DP table&#xA;    for (int i = 2; i &lt;= n; i++) {&#xA;        for (int j = 2; j &lt;= min(i, k + 1); j++) {&#xA;            // Try all possible positions for the last multiplication&#xA;            for (int m = j - 1; m &lt; i; m++) {&#xA;                long long product = dpm  getNumber(s, m, i - 1);&#xA;                dpi = max(dpi, product);&#xA;            }&#xA;        }&#xA;    }&#xA;    &#xA;    return dpn;&#xA;}&#xA;&#xA;int main() {&#xA;    int n, k;&#xA;    string s;&#xA;    &#xA;    cin     n     k;&#xA;    cin     s;&#xA;    &#xA;    cout &lt;&lt; solve(s, n, k) &lt;&lt; endl;&#xA;    &#xA;    return 0;&#xA;}&#xA;&#xA;Python Solution:&#xA;&#xA;def getnumber(s, start, end):&#xA;    &#34;&#34;&#34;Extract number from string s from index start to end (inclusive)&#34;&#34;&#34;&#xA;    return int(s[start:end+1])&#xA;&#xA;def solve(s, n, k):&#xA;    &#34;&#34;&#34;&#xA;    Find maximum product by placing k multiplication operators in string s&#xA;    dpi = maximum product using first i digits with j-1 multiplications&#xA;    &#34;&#34;&#34;&#xA;    # Initialize DP table&#xA;    dp = [[0]  (k + 2) for  in range(n + 1)]&#xA;    &#xA;    # Base case: no multiplication, just the number itself&#xA;    for i in range(1, n + 1):&#xA;        dpi = getnumber(s, 0, i - 1)&#xA;    &#xA;    # Fill DP table&#xA;    for i in range(2, n + 1):&#xA;        for j in range(2, min(i, k + 1) + 1):&#xA;            # Try all possible positions for the last multiplication&#xA;            for m in range(j - 1, i):&#xA;                product = dpm  getnumber(s, m, i - 1)&#xA;                dpi = max(dpi, product)&#xA;    &#xA;    return dpn&#xA;&#xA;def main():&#xA;    n, k = map(int, input().split())&#xA;    s = input().strip()&#xA;    &#xA;    print(solve(s, n, k))&#xA;&#xA;if name == &#34;main&#34;:&#xA;    main()&#xA;&#xA;How it works:&#xA;&#xA;Dynamic Programming Approach: &#xA;   dpi represents the maximum product using the first i digits with j-1 multiplication operators&#xA;   &#xA;Base Case: &#xA;   When j=1 (no multiplications), it&#39;s just the number formed by the first i digits&#xA;&#xA;Recurrence Relation:&#xA;   For each position i and number of parts j, we try all possible positions m for the last multiplication&#xA;   The product is: dpm * numberfrompositionmto_i&#xA;&#xA;Time Complexity: O(n²×k)&#xA;Space Complexity: O(n×k)&#xA;&#xA;For the example input:&#xA;Input: n=4, k=2, s=&#34;1231&#34;&#xA;The algorithm will find that splitting as 1×2×31 gives maximum product = 62&#xA;&#xA;]]&gt;</description>
      <content:encoded><![CDATA[<p>[](<a href="https://pad.degrowth.net/uploads/0ab40389-e49d-4bb1-83e7-aa98b84e0e51.png" rel="nofollow">https://pad.degrowth.net/uploads/0ab40389-e49d-4bb1-83e7-aa98b84e0e51.png</a>)</p>

<p><a href="https://i.postimg.cc/gcxZf342/mul.png" rel="nofollow">mul.png</a></p>

<p>Find the maximum product when placing multiplication operators in a number string.</p>

<h2 id="problem-analysis" id="problem-analysis">Problem Analysis:</h2>

<p>Given:
– A string of N digits (6 ≤ N ≤ 40)
– Need to place K multiplication operators (1 ≤ K ≤ 6)
– Find the maximum product possible</p>

<p>For example, with string “1231” and K=2:
– We need to split it into 3 parts (K+1 parts)
– Find the arrangement that gives maximum product</p>



<h2 id="c-solution" id="c-solution">C++ Solution:</h2>

<pre><code class="language-cpp">#include &lt;iostream&gt;
#include &lt;string&gt;
#include &lt;vector&gt;
#include &lt;algorithm&gt;
#include &lt;climits&gt;
using namespace std;

long long getNumber(const string&amp; s, int start, int end) {
    long long num = 0;
    for (int i = start; i &lt;= end; i++) {
        num = num * 10 + (s[i] - &#39;0&#39;);
    }
    return num;
}

long long solve(const string&amp; s, int n, int k) {
    // dp[i][j] = maximum product using first i digits with j multiplications
    vector&lt;vector&lt;long long&gt;&gt; dp(n + 1, vector&lt;long long&gt;(k + 2, 0));
    
    // Initialize: no multiplication, just the number itself
    for (int i = 1; i &lt;= n; i++) {
        dp[i][1] = getNumber(s, 0, i - 1);
    }
    
    // Fill the DP table
    for (int i = 2; i &lt;= n; i++) {
        for (int j = 2; j &lt;= min(i, k + 1); j++) {
            // Try all possible positions for the last multiplication
            for (int m = j - 1; m &lt; i; m++) {
                long long product = dp[m][j-1] * getNumber(s, m, i - 1);
                dp[i][j] = max(dp[i][j], product);
            }
        }
    }
    
    return dp[n][k + 1];
}

int main() {
    int n, k;
    string s;
    
    cin &gt;&gt; n &gt;&gt; k;
    cin &gt;&gt; s;
    
    cout &lt;&lt; solve(s, n, k) &lt;&lt; endl;
    
    return 0;
}
</code></pre>

<h2 id="python-solution" id="python-solution">Python Solution:</h2>

<pre><code class="language-python">def get_number(s, start, end):
    &#34;&#34;&#34;Extract number from string s from index start to end (inclusive)&#34;&#34;&#34;
    return int(s[start:end+1])

def solve(s, n, k):
    &#34;&#34;&#34;
    Find maximum product by placing k multiplication operators in string s
    dp[i][j] = maximum product using first i digits with j-1 multiplications
    &#34;&#34;&#34;
    # Initialize DP table
    dp = [[0] * (k + 2) for _ in range(n + 1)]
    
    # Base case: no multiplication, just the number itself
    for i in range(1, n + 1):
        dp[i][1] = get_number(s, 0, i - 1)
    
    # Fill DP table
    for i in range(2, n + 1):
        for j in range(2, min(i, k + 1) + 1):
            # Try all possible positions for the last multiplication
            for m in range(j - 1, i):
                product = dp[m][j-1] * get_number(s, m, i - 1)
                dp[i][j] = max(dp[i][j], product)
    
    return dp[n][k + 1]

def main():
    n, k = map(int, input().split())
    s = input().strip()
    
    print(solve(s, n, k))

if __name__ == &#34;__main__&#34;:
    main()
</code></pre>

<h2 id="how-it-works" id="how-it-works">How it works:</h2>
<ol><li><p><strong>Dynamic Programming Approach</strong>:</p>
<ul><li><code>dp[i][j]</code> represents the maximum product using the first <code>i</code> digits with <code>j-1</code> multiplication operators
<br></li></ul></li>

<li><p><strong>Base Case</strong>:</p>
<ul><li>When j=1 (no multiplications), it&#39;s just the number formed by the first i digits</li></ul></li>

<li><p><strong>Recurrence Relation</strong>:</p>
<ul><li>For each position i and number of parts j, we try all possible positions m for the last multiplication</li>
<li>The product is: <code>dp[m][j-1] * number_from_position_m_to_i</code></li></ul></li>

<li><p><strong>Time Complexity</strong>: O(n²×k)</p></li>

<li><p><strong>Space Complexity</strong>: O(n×k)</p></li></ol>

<p>For the example input:
– Input: n=4, k=2, s=“1231”
– The algorithm will find that splitting as 1×2×31 gives maximum product = 62</p>
]]></content:encoded>
      <guid>https://qua.name/xv01/mul-dp</guid>
      <pubDate>Wed, 27 Aug 2025 10:22:06 +0200</pubDate>
    </item>
    <item>
      <title>RLHF</title>
      <link>https://qua.name/xv01/rlhf</link>
      <description>&lt;![CDATA[https://x.com/xleaps/status/1894560176210149655&#xA;&#xA;Centrifugal governor ≈ the 18-century &#34;human-feedback loop&#34; for steam engines.&#xA;https://en.wikipedia.org/wiki/Centrifugalgovernor&#xA;&#xA;RLHF ≈ the 21-century &#34;human-feedback loop&#34; for language models.&#xA;https://en.wikipedia.org/wiki/Reinforcementlearningfromhuman_feedback&#xA;&#xA;Both solve the same meta-problem:&#xA;&#xA;    Desired output - actual output  -  error signal  -  gain-adjusted correction -  stable useful machine]]&gt;</description>
      <content:encoded><![CDATA[<p><a href="https://x.com/xleaps/status/1894560176210149655" rel="nofollow">https://x.com/xleaps/status/1894560176210149655</a></p>

<p>Centrifugal governor ≈ the 18-century “human-feedback loop” for steam engines.
<a href="https://en.wikipedia.org/wiki/Centrifugal_governor" rel="nofollow">https://en.wikipedia.org/wiki/Centrifugal_governor</a></p>

<p>RLHF ≈ the 21-century “human-feedback loop” for language models.
<a href="https://en.wikipedia.org/wiki/Reinforcement_learning_from_human_feedback" rel="nofollow">https://en.wikipedia.org/wiki/Reinforcement_learning_from_human_feedback</a></p>

<p>Both solve the same meta-problem:</p>

<p>    Desired output – actual output  –&gt;  error signal  –&gt;  gain-adjusted correction –&gt;  stable useful machine</p>
]]></content:encoded>
      <guid>https://qua.name/xv01/rlhf</guid>
      <pubDate>Mon, 04 Aug 2025 12:21:29 +0200</pubDate>
    </item>
  </channel>
</rss>