agent-readings

<div class="slide-frame slide-frame--center">
  <p class="deck-kicker">agent-readings / Multi-Agent Systems</p>
  <h1>CAMEL</h1>
  <p class="deck-subtitle">Role-playing as agent data infrastructure</p>

<div class="deck-meta-strip">
    <div class="deck-meta-pill"><strong>2023</strong><span>NeurIPS 2023</span></div>
    <div class="deck-meta-pill"><strong>8</strong><span>claims</span></div>
    <div class="deck-meta-pill"><strong>8</strong><span>evidence refs</span></div>
    <div class="deck-meta-pill"><strong>paper-read</strong><span>status</span></div>
  </div>
</div>

---

<div class="slide-frame">
  <p class="deck-kicker">The one sentence</p>
  <h2>CAMEL uses role-conditioned agent conversations to turn a vague idea into multi-turn instruction-following data and task progress.</h2>
  <p class="deck-note">CAMEL shows why multi-agent prompting is useful beyond solving a single task: dialogue between role-specialized agents becomes a data-production mechanism for studying, training, and stress-testing agent societies.</p>
</div>

---

<div class="slide-frame">
  <p class="deck-kicker">Before this paper</p>
  <h2>The friction was structural.</h2>
  <div class="apple-card-grid apple-card-grid--three">
    <div class="apple-card"><p>Chat LLMs solve complex tasks only when humans keep steering the conversation with good prompts.</p></div>
<div class="apple-card"><p>Manual prompting requires domain knowledge and does not scale to large studies of cooperative agent behavior.</p></div>
<div class="apple-card"><p>Naive agent-agent chats suffer role flipping, instruction repetition, vague replies, and infinite loops.</p></div>
  </div>
  <div class="deck-callout">
    <span>Key gap</span>
    <p>The paper asks whether role-conditioned LLM agents can autonomously cooperate from a preliminary idea and generate useful conversational data with minimal human intervention.</p>
  </div>
</div>

---

<div class="slide-frame">
  <p class="deck-kicker">Core idea</p>
  <h2>Start from roles and a rough idea, use a task-specifier prompt to sharpen the task, initialize an AI user and AI assistant with symmetric role-preserving...</h2>
  <p class="deck-note">Start from roles and a rough idea, use a task-specifier prompt to sharpen the task, initialize an AI user and AI assistant with symmetric role-preserving prompts, then let them alternate instruction and solution messages until a completion token or other...</p>
  <div class="idea-loop" aria-label="Agent loop">
    <div><strong>Model</strong><span>reason</span></div>
    <div><strong>Runtime</strong><span>act</span></div>
    <div><strong>World</strong><span>observe</span></div>
  </div>
  <div class="step-strip">
    <div class="step-item">
  <span>01</span>
  <p>Generate or accept assistant roles, user roles, and preliminary tasks.</p>
</div>
<div class="step-item">
  <span>02</span>
  <p>Use the task-specifier prompt to produce a more concrete task.</p>
</div>
<div class="step-item">
  <span>03</span>
  <p>Instantiate the AI user and AI assistant with inception prompts that prohibit role flipping and define response format.</p>
</div>
<div class="step-item">
  <span>04</span>
  <p>Run the two-agent loop where the user instructs and the assistant solves.</p>
</div>
  </div>
</div>

---

<div class="slide-frame">
  <p class="deck-kicker">Action grammar</p>
  <h2>The agent is an interface contract.</h2>
  <div class="apple-card-grid apple-card-grid--two">
    <div class="apple-card"><p>Role generation: choose or generate assistant and user roles.</p></div>
<div class="apple-card"><p>Task generation: generate tasks that the roles can cooperatively complete.</p></div>
<div class="apple-card"><p>Task specification: make a preliminary idea concrete in a bounded number of words.</p></div>
<div class="apple-card"><p>Role initialization: provide each agent with its role, the other agent's role, task, protocol, and constraints.</p></div>
  </div>
</div>

---

<div class="slide-frame">
  <p class="deck-kicker">Executable shape</p>
  <h2>The mechanism should fit on one screen.</h2>

```python
roles = generate_roles()
idea = generate_or_receive_idea()
task = task_specifier(roles, idea)

ai_user = LLM(system_prompt=user_inception_prompt(roles, task))
ai_assistant = LLM(system_prompt=assistant_inception_prompt(roles, task))
messages = []

for turn in range(max_messages):
    instruction = ai_user(messages)
    if instruction == "CAMEL_TASK_DONE" or violates_protocol(instruction):
        break
    solution = ai_assistant(messages, instruction)
    if role_flip(solution) or token_limit_reached(solution):
        break
    messages.append({"instruction": instruction, "solution": solution})
```
</div>

---

<div class="slide-frame">
  <p class="deck-kicker">Evidence</p>
  <h2>The proof objects.</h2>
  <div class="proof-grid">
    <div class="proof-card">
  <span>E-role-playing</span>
  <p>The framework starts from a preliminary idea and role assignment; a task-specifier makes the task concrete; the AI user instructs one step at a time and the AI assistant provides...</p>
</div>
<div class="proof-card">
  <span>E-inception</span>
  <p>Inception prompts include a task-specifier prompt plus assistant and user system prompts with role declarations, role-flip prohibitions, instruction/solution formats, task...</p>
</div>
<div class="proof-card">
  <span>E-dataset-scale</span>
  <p>The AI Society generation process creates 50 assistant roles, 50 user roles, and 10 tasks for every role pair, for 25,000 conversations. The paper also generates Code, Math,...</p>
</div>
<div class="proof-card">
  <span>E-failures</span>
  <p>The paper reports four major autonomous cooperation failures: role flipping, assistant repeating instructions, vague flake replies, and infinite loops of meaningless messages.</p>
</div>
  </div>
</div>

---

<div class="slide-frame">
  <p class="deck-kicker">Claim map</p>
  <h2>What the review actually supports.</h2>
  <div class="claim-grid">
    <div class="claim-card">
  <span>C1 · paper-supported</span>
  <p>CAMEL's role-playing framework assigns an AI user to plan and instruct, and an AI assistant to execute instructions and provide solutions.</p>
  <small>E-role-playing</small>
</div>
<div class="claim-card">
  <span>C2 · paper-supported</span>
  <p>Inception prompting front-loads task specification, role assignment, communication protocol, termination, and anti-failure constraints before autonomous...</p>
  <small>E-inception</small>
</div>
<div class="claim-card">
  <span>C3 · paper-supported</span>
  <p>For AI Society, CAMEL generates 50 assistant roles, 50 user roles, and 10 tasks for each role pair, yielding 25,000 conversations.</p>
  <small>E-dataset-scale</small>
</div>
<div class="claim-card">
  <span>C4 · paper-supported</span>
  <p>The paper identifies role flipping, repeated instructions, flake replies, and infinite loops as core failure modes for autonomous agent-agent dialogue.</p>
  <small>E-failures</small>
</div>
  </div>
</div>

---

<div class="slide-frame">
  <p class="deck-kicker">Agent infrastructure</p>
  <h2>What changes if you build systems.</h2>
  <div class="apple-card-grid apple-card-grid--two">
    <div class="apple-card"><p>Role prompts are protocol definitions. CAMEL's guardrails are essentially a communication contract between agents.</p></div>
<div class="apple-card"><p>A multi-agent dialogue loop needs termination policy before it needs more agents. Role flipping and infinite loops are runtime bugs.</p></div>
<div class="apple-card"><p>Synthetic agent conversations are a data engine. CAMEL's biggest infrastructure lesson is that agent-agent traces can become training and evaluation...</p></div>
<div class="apple-card"><p>Separate planner and executor roles. The AI user decides what to ask next; the AI assistant executes the instruction.</p></div>
  </div>
</div>

---

<div class="slide-frame">
  <p class="deck-kicker">Caveats</p>
  <h2>What this does not prove.</h2>
  <div class="apple-card-grid apple-card-grid--two">
    <div class="apple-card"><p>The framework is prompt-engineered and does not formally guarantee role stability or task success.</p></div>
<div class="apple-card"><p>The evaluations rely heavily on GPT-4 and human preference judgments.</p></div>
<div class="apple-card"><p>CAMEL's generated datasets inherit biases, omissions, and safety issues from the models used to generate them.</p></div>
<div class="apple-card"><p>The role-playing setup focuses on cooperative assistant-user settings, not adversarial or fully open multi-agent economies.</p></div>
  </div>
</div>

---

<div class="slide-frame">
  <p class="deck-kicker">Run it</p>
  <h2>No PoC yet.</h2>
  <p class="deck-note">No PoC yet - contributions welcome. A compact PoC would implement task specification plus a two-agent instruction/solution loop with role-flip detection and CAMEL_TASK_DONE termination over a local or mocked LLM.</p>
  <div class="reference-list">
    <ul><li>Paper: <a href="https://arxiv.org/abs/2303.17760">CAMEL: Communicative Agents for Mind Exploration of Large Language Model Society</a></li><li>Project/code: <a href="https://github.com/camel-ai/camel">https://github.com/camel-ai/camel</a></li><li>contrast: <a href="https://arxiv.org/abs/2308.08155">AutoGen</a></li><li>parallel: <a href="https://arxiv.org/abs/2305.14325">Multiagent Debate</a></li><li>connects-to: <a href="https://arxiv.org/abs/2304.03442">Generative Agents</a></li></ul>
  </div>
</div>

Note: This deck is synthesized from `data/reviews/camel-li-2023.json`. Update the review record, then run `bun run build`.