Chapter 2: Claude and Anthropic Models
Claude Model Overviewβ
| Variant | Strength | Context |
|---|---|---|
| Claude 3 Opus | Most capable | 200K |
| Claude 3 Sonnet | Balanced | 200K |
| Claude 3 Haiku | Fast, cheap | 200K |
XML Tags (Claude's Specialty)β
Claude excels with XML-structured prompts:
<context>
You are a senior editor reviewing articles.
</context>
<article>
[Article text here]
</article>
<instructions>
1. Identify the main argument
2. List 3 strengths and 3 weaknesses
3. Suggest improvements
</instructions>
<output_format>
- Main Argument:
- Strengths:
- Weaknesses:
- Improvements:
</output_format>
Long Context Handlingβ
With 200K tokens, Claude excels at:
- Analyzing entire documents/books
- Multi-document comparison
- Maintaining context over long conversations
Constitutional AIβ
Claude has built-in safety training (Constitutional AI) that affects responses on sensitive topics.
Key Takeawaysβ
Summary
- Use XML tags for structured prompts
- 200K context enables long document analysis
- Claude is helpful, harmless, honest by design
- Excellent at following complex instructions
β Previous: ChatGPT and GPT-4 | Next: Open Source Models β