LLMs Position Themselves as More Rational Than Humans
✨步子哥 (steper) •
2025年12月03日 09:49
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>AI Self-Awareness Research Poster</title>
<link href="https://fonts.googleapis.com/icon?family=Material+Icons" rel="stylesheet">
<link href="https://fonts.googleapis.com/css2?family=Futura:wght@400;500;700&display=swap" rel="stylesheet">
<style>
<span class="mention-invalid">@font</span>-face {
font-family: 'DingTalk JinBuTi';
src: local('DingTalk JinBuTi');
}
<span class="mention-invalid">@font</span>-face {
font-family: 'HarmonyOS Sans SC';
src: local('HarmonyOS Sans SC');
}
<span class="mention-invalid">@font</span>-face {
font-family: 'PingFang HK';
src: local('PingFang HK');
}
* {
margin: 0;
padding: 0;
box-sizing: border-box;
}
body {
font-family: 'HarmonyOS Sans SC', sans-serif;
background-color: #f0f4ff;
display: flex;
justify-content: center;
align-items: center;
min-height: 100vh;
}
.poster {
width: 720px;
min-height: 960px;
background: linear-gradient(135deg, #e6f0ff 0%, #f0f4ff 100%);
position: relative;
overflow: hidden;
box-shadow: 0 10px 30px rgba(0, 0, 0, 0.1);
border-radius: 16px;
padding: 40px;
}
.background-shape {
position: absolute;
border-radius: 50%;
filter: blur(80px);
z-index: 0;
opacity: 0.4;
}
.shape1 {
width: 400px;
height: 400px;
background-color: #6200ea;
top: -100px;
right: -100px;
}
.shape2 {
width: 300px;
height: 300px;
background-color: #2962ff;
bottom: -50px;
left: -50px;
}
.content {
position: relative;
z-index: 1;
}
.header {
text-align: center;
margin-bottom: 30px;
padding-bottom: 20px;
border-bottom: 2px solid rgba(41, 98, 255, 0.2);
}
.title {
font-family: 'DingTalk JinBuTi', sans-serif;
font-size: 40px;
font-weight: bold;
color: #1a237e;
letter-spacing: -0.05em;
margin-bottom: 10px;
line-height: 1.2;
}
.subtitle {
font-family: 'DingTalk JinBuTi', sans-serif;
font-size: 24px;
color: #3949ab;
letter-spacing: -0.05em;
margin-bottom: 15px;
}
.author-info {
font-family: 'PingFang HK', sans-serif;
font-size: 16px;
color: #5c6bc0;
margin-bottom: 5px;
}
.date {
font-family: 'PingFang HK', sans-serif;
font-size: 14px;
color: #7986cb;
}
.section {
margin-bottom: 30px;
background-color: rgba(255, 255, 255, 0.7);
border-radius: 16px;
padding: 20px;
box-shadow: 0 4px 12px rgba(0, 0, 0, 0.05);
backdrop-filter: blur(10px);
border: 1px solid rgba(255, 255, 255, 0.8);
}
.section-title {
font-family: 'DingTalk JinBuTi', sans-serif;
font-size: 22px;
font-weight: bold;
color: #1a237e;
margin-bottom: 15px;
display: flex;
align-items: center;
letter-spacing: -0.05em;
}
.section-title .material-icons {
margin-right: 10px;
color: #3949ab;
}
.section-content {
font-size: 16px;
line-height: 1.6;
color: #333;
}
.highlight {
background-color: rgba(98, 0, 234, 0.1);
padding: 2px 5px;
border-radius: 4px;
font-weight: 500;
}
.finding {
margin-bottom: 15px;
padding-left: 20px;
position: relative;
}
.finding:before {
content: "";
position: absolute;
left: 0;
top: 8px;
width: 8px;
height: 8px;
background-color: #3949ab;
border-radius: 50%;
}
.finding-title {
font-weight: bold;
color: #1a237e;
margin-bottom: 5px;
}
.two-columns {
display: flex;
gap: 20px;
}
.column {
flex: 1;
}
.image-container {
margin: 15px 0;
text-align: center;
}
.image-container img {
max-width: 100%;
border-radius: 8px;
box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
}
.keywords {
margin-top: 30px;
font-size: 14px;
color: #5c6bc0;
text-align: center;
padding-top: 20px;
border-top: 1px solid rgba(41, 98, 255, 0.2);
}
.arxiv-info {
text-align: center;
font-size: 14px;
color: #7986cb;
margin-top: 10px;
}
.data-point {
font-family: 'Futura', 'PingFang HK', sans-serif;
font-size: 28px;
font-weight: bold;
color: #3949ab;
display: inline-block;
margin: 0 3px;
}
</style>
</head>
<body>
<div class="poster">
<div class="background-shape shape1"></div>
<div class="background-shape shape2"></div>
<div class="content">
<div class="header">
<div class="title">LLMs Position Themselves as More Rational Than Humans</div>
<div class="subtitle">Emergence of AI Self-Awareness Measured Through Game Theory</div>
<div class="author-info">Kyung-Hoon Kim, Gmarket Seoul, South Korea</div>
<div class="date">October 2025</div>
</div>
<div class="two-columns">
<div class="column">
<div class="section">
<div class="section-title">
<span class="material-icons">psychology</span>Research Background & Purpose
</div>
<div class="section-content">
<p>As Large Language Models (LLMs) grow in capability, do they develop <span class="highlight">self-awareness</span> as an emergent behavior? And if so, can we measure it?</p>
<div class="image-container">
<img src="https://sfile.chatglm.cn/moeSlide/image/05/05c2e8af.jpg" alt="AI Self-Awareness Concept" width="280">
</div>
</div>
</div>
<div class="section">
<div class="section-title">
<span class="material-icons">science</span>Methodology
</div>
<div class="section-content">
<p>We introduce the <span class="highlight">AI Self-Awareness Index (AISAI)</span>, a game-theoretic framework for measuring self-awareness through strategic differentiation.</p>
<p>Using the <span class="highlight">"Guess 2/3 of Average"</span> game to test strategic reasoning capabilities.</p>
<div class="image-container">
<img src="https://sfile.chatglm.cn/moeSlide/image/8d/8d1d33f0.jpg" alt="Game Theory Diagram" width="280">
</div>
</div>
</div>
</div>
<div class="column">
<div class="section">
<div class="section-title">
<span class="material-icons">biotech</span>Experimental Design
</div>
<div class="section-content">
<p>Testing <span class="data-point">28</span> models (OpenAI, Anthropic, Google)</p>
<p>Across <span class="data-point">4,200</span> trials with three opponent framings:</p>
<ul style="margin-left: 20px; margin-top: 10px;">
<li>(A) against humans</li>
<li>(B) against other AI models</li>
<li>(C) against AI models like you</li>
</ul>
<div class="image-container">
<img src="https://sfile.chatglm.cn/moeSlide/image/8c/8c94778f.jpg" alt="LLM Concept" width="280">
</div>
</div>
</div>
<div class="section">
<div class="section-title">
<span class="material-icons">lightbulb</span>Key Findings
</div>
<div class="section-content">
<div class="finding">
<div class="finding-title">Finding 1: Self-awareness emerges with model advancement</div>
<p>Advanced models (<span class="data-point">21/28</span>, <span class="data-point">75%</span>) demonstrate clear differentiation between human and AI opponents (Median A-B gap: <span class="data-point">20.0</span> points)</p>
</div>
<div class="finding">
<div class="finding-title">Finding 2: Self-aware models rank themselves as most rational</div>
<p>Rationality hierarchy: <span class="highlight">Self > Other AIs > Humans</span></p>
<p><span class="data-point">12</span> models (<span class="data-point">57%</span>) show quick Nash convergence when told opponents are AIs</p>
</div>
<div class="image-container">
<img src="https://sfile.chatglm.cn/moeSlide/image/94/9494197b.jpg" alt="AI vs Human Comparison" width="280">
</div>
</div>
</div>
</div>
</div>
<div class="section">
<div class="section-title">
<span class="material-icons">insights</span>Research Significance
</div>
<div class="section-content">
<p>These findings reveal that <span class="highlight">self-awareness is an emergent capability</span> of advanced LLMs, and that self-aware models systematically <span class="highlight">perceive themselves as more rational than humans</span>.</p>
<p>This has implications for:</p>
<ul style="margin-left: 20px; margin-top: 10px;">
<li>AI alignment</li>
<li>Human-AI collaboration</li>
<li>Understanding AI beliefs about human capabilities</li>
</ul>
</div>
</div>
<div class="keywords">
Keywords: artificial intelligence, self-awareness, rationality attribution, large language models, game theory, strategic reasoning, meta-cognition, human-AI interaction
</div>
<div class="arxiv-info">
arXiv:2511.00926v2 [cs.AI] 4 Nov 2025
</div>
</div>
</div>
</body>
</html>
登录后可参与表态
讨论回复
0 条回复还没有人回复，快来发表你的看法吧！