Chapter 2: Claude (Anthropic)
Anthropic's Mission
Founded by former OpenAI researchers, Anthropic focuses on AI safety and alignment.
Constitutional AI
Instead of human labeling for every case, Claude is trained with a "constitution" of principles:
1. Be helpful, harmless, and honest
2. Avoid deception
3. Respect human autonomy
4. Support human oversight
Claude Models
| Model | Context | Best For |
|---|---|---|
| Claude 3 Haiku | 200K | Fast, cost-effective |
| Claude 3 Sonnet | 200K | Balanced performance |
| Claude 3 Opus | 200K | Complex tasks |
Differentiators
- Long context: 200K tokens (entire codebases)
- Safety: Strong refusal on harmful requests
- Honesty: Acknowledges uncertainty
- Helpfulness: Detailed, thoughtful responses
Key Takeaways
Summary
- Constitutional AI provides principled alignment
- Claude 3 offers excellent long-context capabilities
- Safety-first approach influences all design decisions