Loading...
正在加载...
请稍候

Google DeepMind Intrinsic Self-Critique “诗人” vs “计算器”

✨步子哥 (steper) 2026年02月11日 12:23
<!DOCTYPE html> <html lang="zh-CN"> <head> <meta charset="UTF-8"> <meta name="viewport" content="width=device-width, initial-scale=1.0"> <title>当“诗人”被迫拿起算盘 - Google DeepMind</title> <link href="https://fonts.googleapis.com/icon?family=Material+Icons" rel="stylesheet"> <link href="https://fonts.googleapis.com/css2?family=Roboto+Mono:wght@400;700&family=Ma+Shan+Zheng&family=Noto+Serif+SC:wght@400;700&display=swap" rel="stylesheet"> <style> :root { --primary-color: #2E2A6B; /* Deep Purple */ --secondary-color: #424242; /* Dark Grey */ --accent-color: #FFD700; /* Gold for contrast */ --bg-color: #F5F5F7; --card-bg: #FFFFFF; --text-main: #1A1A1A; --text-light: #757575; } * { box-sizing: border-box; margin: 0; padding: 0; } body { font-family: "HarmonyOS Sans SC", "Noto Sans SC", sans-serif; background-color: #e0e0e0; display: flex; justify-content: center; min-height: 100vh; } .poster-container { width: 720px; min-height: 960px; background-color: var(--bg-color); display: flex; flex-direction: column; overflow: hidden; position: relative; box-shadow: 0 10px 30px rgba(0,0,0,0.15); } /* Background Decoration */ .bg-shape { position: absolute; border-radius: 50%; filter: blur(60px); z-index: 0; } .bg-shape-1 { width: 400px; height: 400px; background: rgba(46, 42, 107, 0.15); top: -100px; right: -100px; } .bg-shape-2 { width: 300px; height: 300px; background: rgba(255, 215, 0, 0.1); bottom: 100px; left: -50px; } /* Header */ header { padding: 40px 40px 20px 40px; z-index: 1; position: relative; } .main-title { font-size: 48px; line-height: 1.1; color: var(--primary-color); margin-bottom: 10px; font-weight: 900; } .title-poet { font-family: 'Swei B2 Serif CJKtc Bold', 'Noto Serif SC', serif; font-style: italic; color: #5E35B1; } .title-calc { font-family: 'Futura', 'Roboto Mono', monospace; color: var(--secondary-color); } .subtitle { font-size: 20px; color: var(--text-light); font-weight: 500; display: flex; align-items: center; gap: 8px; } .tag { background: var(--secondary-color); color: white; padding: 4px 10px; border-radius: 4px; font-size: 14px; font-family: 'Roboto Mono', monospace; text-transform: uppercase; } /* Content Layout */ .content { flex: 1; padding: 20px 40px 40px 40px; display: flex; flex-direction: column; gap: 25px; z-index: 1; } /* Cards */ .card { background: var(--card-bg); border-radius: 16px; padding: 24px; box-shadow: 0 4px 12px rgba(0,0,0,0.05); position: relative; overflow: hidden; transition: transform 0.2s; } .card-title { font-size: 22px; font-weight: 700; color: var(--primary-color); margin-bottom: 16px; display: flex; align-items: center; gap: 10px; border-left: 4px solid var(--accent-color); padding-left: 12px; } .card-title .material-icons { color: var(--accent-color); } /* Section 1: The Metaphor */ .metaphor-section { display: flex; gap: 20px; align-items: stretch; } .metaphor-visual { flex: 0 0 40%; position: relative; border-radius: 12px; overflow: hidden; } .metaphor-visual img { width: 100%; height: 100%; object-fit: cover; } .metaphor-text { flex: 1; display: flex; flex-direction: column; justify-content: center; } .vs-badge { position: absolute; top: 50%; left: 50%; transform: translate(-50%, -50%); background: var(--accent-color); color: var(--secondary-color); width: 32px; height: 32px; border-radius: 50%; display: flex; align-items: center; justify-content: center; font-weight: bold; font-size: 14px; z-index: 2; box-shadow: 0 2px 8px rgba(0,0,0,0.2); } .p-text { font-size: 15px; line-height: 1.6; color: var(--text-main); margin-bottom: 8px; } .highlight { color: var(--primary-color); font-weight: 700; } /* Section 2: PDDL & Mechanism Grid */ .grid-2-col { display: grid; grid-template-columns: 1fr 1fr; gap: 20px; } .pddl-box { background: #F3F4F6; padding: 15px; border-radius: 8px; border: 1px dashed var(--primary-color); } .code-snippet { font-family: 'Roboto Mono', monospace; font-size: 12px; color: #555; margin-top: 8px; background: white; padding: 8px; border-radius: 4px; } /* Section 3: Results */ .result-container { display: flex; align-items: center; justify-content: space-around; background: linear-gradient(135deg, #2E2A6B 0%, #5E35B1 100%); color: white; } .result-item { text-align: center; } .big-number { font-size: 48px; font-weight: 900; color: var(--accent-color); font-family: 'Futura', sans-serif; } .arrow-icon { font-size: 36px; color: rgba(255,255,255,0.7); } .result-label { font-size: 14px; opacity: 0.9; margin-top: 4px; } /* Section 4: Philosophy */ .footer-card { border: none; background: transparent; box-shadow: none; padding: 0; } .footer-content { display: flex; align-items: center; background: white; border-radius: 12px; padding: 20px; border-left: 8px solid var(--secondary-color); } .footer-icon { font-size: 40px; color: var(--secondary-color); margin-right: 20px; } /* Illustrations */ .icon-group { display: flex; justify-content: flex-end; gap: 10px; margin-bottom: 10px; } .small-icon { width: 24px; height: 24px; background: #EEE; border-radius: 4px; display: flex; align-items: center; justify-content: center; font-size: 14px; color: #666; } </style> </head> <body> <div class="poster-container"> <!-- Background Elements --> <div class="bg-shape bg-shape-1"></div> <div class="bg-shape bg-shape-2"></div> <header> <div class="subtitle"> <span class="tag">Google DeepMind</span> <span>Intrinsic Self-Critique</span> </div> <h1 class="main-title"> <span class="title-poet">“诗人”</span> vs <span class="title-calc">“计算器”</span> </h1> <div class="subtitle" style="font-size: 16px; margin-top: 5px;"> <i class="material-icons" style="font-size: 18px;">psychology</i> 强迫 LLM 进行逻辑推理的暴力美学 </div> </header> <div class="content"> <!-- Section 1: The Metaphor --> <div class="card"> <div class="metaphor-section"> <div class="metaphor-visual"> <div class="vs-badge">VS</div> <img src="https://sfile.chatglm.cn/image/78/784dbc3d.jpg" alt="Brain Hemispheres"> </div> <div class="metaphor-text"> <p class="p-text"> <span class="highlight">诗人 (LLM):</span> 基于概率的生成,天马行空,充满创造力,但缺乏严谨的逻辑确定性。 </p> <p class="p-text"> <span class="highlight">计算器 (Logic):</span> 需要确定性的规则与结果,容不得半点模糊。 </p> <p class="p-text" style="margin-top: 10px; font-size: 14px; color: var(--text-light);"> Google DeepMind 试图打破这一界限,让“诗人”学会“计算”。 </p> </div> </div> </div> <!-- Section 2: PDDL Sandbox & Method --> <div class="grid-2-col"> <div class="card"> <div class="card-title"> <i class="material-icons">security</i> PDDL 沙盒 </div> <p class="p-text"> 利用 <strong>PDDL 规则</strong> 构建“物理引擎”,在混沌的语言世界中划定边界。 </p> <div class="pddl-box"> <div class="icon-group"> <div class="small-icon"><i class="material-icons" style="font-size:14px">block</i></div> <div class="small-icon"><i class="material-icons" style="font-size:14px">rule</i></div> </div> <div class="code-snippet"> (:action pick-up<br> &nbsp;&nbsp;:precondition (and (clear ?ob) ...)<br> &nbsp;&nbsp;:effect (holding ?ob)) </div> </div> <p class="p-text" style="margin-top:10px; font-size:13px;"> 强制 AI 在规则内行动,而非自由联想。 </p> </div> <div class="card"> <div class="card-title"> <i class="material-icons">fact_check</i> 内在自我批判 </div> <p class="p-text"> 让 AI 扮演“多重人格”,通过 <strong>自我审查</strong> 和投票机制进行迭代优化。 </p> <ul style="padding-left: 20px; color: var(--text-light); font-size: 14px; margin-top: 10px;"> <li style="margin-bottom: 5px;">生成计划</li> <li style="margin-bottom: 5px;">自我批判 (Verify)</li> <li>修正完善</li> </ul> <div style="margin-top: auto; text-align: center;"> <img src="https://sfile.chatglm.cn/image/cb/cb64269a.jpg" style="width: 100%; height: 80px; object-fit: cover; border-radius: 8px; opacity: 0.8;" alt="AI Thinking"> </div> </div> </div> <!-- Section 3: Results --> <div class="card result-container"> <div class="result-item"> <div class="result-label">基线准确率</div> <div class="big-number">49%</div> </div> <i class="material-icons arrow-icon">arrow_forward</i> <div class="result-item"> <div class="result-label">自我批判后</div> <div class="big-number">~90%</div> </div> <div style="width: 1px; height: 40px; background: rgba(255,255,255,0.3); margin: 0 10px;"></div> <div class="result-item" style="text-align: left;"> <div class="result-label">关键提升</div> <div style="font-weight: 700; font-size: 16px; color: var(--accent-color);">无需外部验证</div> <div style="font-size: 12px; opacity: 0.8;">完全依靠模型自身反思</div> </div> </div> <!-- Section 4: Conclusion --> <div class="card footer-card"> <div class="footer-content"> <i class="material-icons footer-icon">help_outline</i> <div> <div style="font-weight: 700; font-size: 18px; color: var(--text-main); margin-bottom: 4px;">暴力美学 vs. 效率悖论</div> <p style="font-size: 14px; color: var(--text-light);"> 巨量算力换取简单逻辑,是通往 AGI 的必经之路,还是对资源的挥霍?<br> <strong>神经符号 AI (Neuro-symbolic AI)</strong> 或许是未来的答案。 </p> </div> </div> </div> </div> </div> </body> </html>

讨论回复

0 条回复

还没有人回复,快来发表你的看法吧!