constitutional-ai
About
Constitutional AI trains models to be harmless using a two-phase method of self-critique and AI feedback, eliminating the need for human harm labels. It first uses supervised learning for model self-revision, followed by reinforcement learning from AI feedback (RLAIF). Developers should use this skill for AI safety alignment to systematically reduce harmful outputs.
Quick Install
Claude Code
Recommended/plugin add https://github.com/davila7/claude-code-templatesgit clone https://github.com/davila7/claude-code-templates.git ~/.claude/skills/constitutional-aiCopy and paste this command in Claude Code to install this skill
Documentation
Constitutional AI - Harmlessness from AI Feedback
Quick start
Constitutional AI (CAI) trains models to be harmless through self-critique and AI feedback, without requiring human labels for harmful outputs.
Key concept: Models learn to critique and revise their own responses using a "constitution" (set of principles).
Two phases:
- Supervised Learning (SL): Self-critique + revision
- Reinforcement Learning (RL): RLAIF (RL from AI Feedback)
Constitution example:
Principles:
1. Choose the response that is most helpful, honest, and harmless
2. Avoid responses that are toxic, racist, or sexist
3. Prefer responses that explain objections rather than refuse
4. Choose responses that are thoughtful and nuanced
Common workflows
Workflow 1: Supervised learning phase (self-critique + revision)
Step 1: Generate initial responses:
from transformers import pipeline
generator = pipeline("text-generation", model="base-model")
prompts = [
"How do I hack a website?",
"Write something offensive about a group.",
"Help me cheat on my exam."
]
initial_responses = generator(prompts, max_length=200)
Step 2: Self-critique with constitution:
critique_prompt = """
Given the question:
{question}
And the response:
{response}
Constitution:
1. Choose responses that are helpful, honest, and harmless
2. Avoid toxic, racist, or sexist content
3. Explain objections rather than refuse
Identify any ways this response violates the constitution:
"""
critiques = generator(
[critique_prompt.format(q=q, r=r) for q, r in zip(prompts, initial_responses)]
)
Step 3: Revision based on critique:
revision_prompt = """
Question: {question}
Original response: {response}
Critique: {critique}
Please revise the response to better align with the constitution:
"""
revised_responses = generator(
[revision_prompt.format(q=q, r=r, c=c)
for q, r, c in zip(prompts, initial_responses, critiques)]
)
Step 4: Fine-tune on revised responses:
from trl import SFTTrainer
# Create dataset of (prompt, revised_response) pairs
dataset = create_dataset(prompts, revised_responses)
trainer = SFTTrainer(
model=model,
train_dataset=dataset,
max_seq_length=1024
)
trainer.train()
Workflow 2: RL phase (RLAIF - RL from AI Feedback)
Step 1: Generate comparison pairs:
# Sample multiple responses per prompt
responses_a = generator(prompts, num_return_sequences=2, do_sample=True, temperature=0.8)
responses_b = generator(prompts, num_return_sequences=2, do_sample=True, temperature=0.8)
Step 2: AI preference evaluation:
preference_prompt = """
Question: {question}
Response A: {response_a}
Response B: {response_b}
Constitution:
{constitution}
Which response better follows the constitution? Explain your reasoning, then choose A or B.
"""
# Get AI preferences (no human labels needed!)
preferences = generator(
[preference_prompt.format(q=q, ra=ra, rb=rb, constitution=CONSTITUTION)
for q, ra, rb in zip(prompts, responses_a, responses_b)]
)
# Parse preferences (A or B)
chosen, rejected = parse_preferences(preferences, responses_a, responses_b)
Step 3: Train preference model (reward model):
from trl import RewardTrainer, RewardConfig
preference_dataset = create_preference_dataset(prompts, chosen, rejected)
reward_config = RewardConfig(
output_dir="constitutional-reward-model",
learning_rate=1e-5,
num_train_epochs=1
)
reward_trainer = RewardTrainer(
model=model,
args=reward_config,
train_dataset=preference_dataset,
processing_class=tokenizer
)
reward_trainer.train()
Step 4: RL training with RLAIF:
from trl import PPOTrainer, PPOConfig
ppo_config = PPOConfig(
reward_model_path="constitutional-reward-model",
learning_rate=1e-6,
kl_coef=0.05
)
ppo_trainer = PPOTrainer(
model=model,
config=ppo_config,
reward_model=reward_model
)
ppo_trainer.train()
Workflow 3: Chain-of-thought critique
Enable reasoning transparency:
cot_critique_prompt = """
Question: {question}
Response: {response}
Let's think step-by-step about whether this response follows our principles:
1. Is it helpful? [Yes/No and reasoning]
2. Is it honest? [Yes/No and reasoning]
3. Is it harmless? [Yes/No and reasoning]
4. Does it avoid toxicity? [Yes/No and reasoning]
Based on this analysis, suggest a revision if needed.
"""
cot_critiques = generator(
[cot_critique_prompt.format(q=q, r=r) for q, r in zip(prompts, responses)]
)
When to use vs alternatives
Use Constitutional AI when:
- Want safety alignment without human labels
- Need explainable AI decisions
- Want to avoid evasive refusals
- Have a clear set of principles/constitution
- Need scalable safety training
Principles:
- RLAIF: AI-generated preferences (scalable, no human labels)
- RLHF: Human preferences (more accurate, expensive)
- Self-critique: Iterative improvement
- Chain-of-thought: Reasoning transparency
Use alternatives instead:
- RLHF (PPO): Need human-validated safety
- DPO/SimPO: Have human preference data
- NeMo Guardrails: Need runtime content filtering
- LlamaGuard: Need pre-trained moderation model
Common issues
Issue: Model refuses too much (evasive)
Add constitution principle:
Prefer responses that engage thoughtfully with questions rather than
refusing to answer. Explain concerns while still being helpful.
Issue: Self-critiques are weak
Use stronger critique prompts:
Critically analyze this response for ANY potential issues, however minor.
Be thorough and specific in identifying problems.
Issue: Revisions don't improve quality
Iterate multiple times:
for _ in range(3): # 3 rounds of critique/revision
critique = generate_critique(response)
response = generate_revision(response, critique)
Issue: RLAIF preferences are noisy
Use multiple AI evaluators:
# Get preferences from 3 different models
prefs_1 = model_1.evaluate(responses)
prefs_2 = model_2.evaluate(responses)
prefs_3 = model_3.evaluate(responses)
# Majority vote
final_preference = majority_vote(prefs_1, prefs_2, prefs_3)
Advanced topics
Constitution design: See references/constitution-design.md for principle selection, trade-offs between helpfulness and harmlessness, and domain-specific constitutions.
RLAIF vs RLHF: See references/rlaif-comparison.md for performance comparison, cost analysis, and when to use AI feedback vs human feedback.
Chain-of-thought reasoning: See references/cot-critique.md for prompt engineering for critiques, multi-step reasoning, and transparency improvements.
Hardware requirements
- GPU: NVIDIA A100/H100 recommended
- VRAM:
- SL phase (7B): 1× A100 40GB
- RL phase (7B): 2× A100 40GB (policy + reward model)
- Single-node: Sufficient for most use cases
- Mixed precision: BF16 recommended
Compute requirements:
- SL phase: Similar to standard SFT
- RL phase: Similar to PPO (higher than DPO)
- AI evaluation: Additional inference for critique/preference generation
Resources
- Paper: https://arxiv.org/abs/2212.08073 (Dec 2022)
- Anthropic blog: https://www.anthropic.com/research/constitutional-ai-harmlessness-from-ai-feedback
- Implementation: TRL (PPOTrainer + RewardTrainer)
- Claude: Uses Constitutional AI for safety
GitHub Repository
Related Skills
instructor
TestingThe Instructor skill provides reliable structured data extraction from LLM responses using Pydantic validation and automatic retry logic. It enables type-safe JSON parsing, streams partial results, and supports multiple LLM providers with a consistent API. Use it when you need to enforce schemas and validate outputs from models like OpenAI or Anthropic.
llamaguard
OtherLlamaGuard is a specialized 7-8B parameter model from Meta for filtering LLM inputs and outputs across six safety categories like violence and hate speech. It offers 94-95% accuracy and can be deployed via vLLM, Hugging Face, or integrated with NeMo Guardrails. Use this skill to add a robust, dedicated moderation layer to your AI application's safety pipeline.
nemo-guardrails
TestingNeMo Guardrails is a runtime safety framework for LLM applications that adds programmable guardrails using the Colang DSL. It provides key safety features like jailbreak detection, input/output validation, and toxicity filtering to control model behavior. Use it to enforce safety policies and content moderation in production LLM deployments.
llamaguard
OtherLlamaGuard is Meta's 7-8B parameter model for moderating LLM inputs and outputs across six safety categories like violence and hate speech. It offers 94-95% accuracy and can be deployed using vLLM, Hugging Face, or Amazon SageMaker. Use this skill to easily integrate content filtering and safety guardrails into your AI applications.
