constitutional-ai

Metode Anthropic untuk melatih AI yang tidak berbahaya melalui perbaikan diri. Pendekatan dua fase - pembelajaran terawasi dengan kritik diri/revisi, kemudian RLAIF (RL dari…

npx skills add https://github.com/firecrawl/ai-research-skills --skill constitutional-ai

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:

  1. Supervised Learning (SL): Self-critique + revision
  2. 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

Lebih banyak skill dari firecrawl

oracle
firecrawl
Praktik terbaik dalam menggunakan CLI oracle (penggabungan prompt dan file, mesin, sesi, dan pola lampiran file).
official
firecrawl-monitor
firecrawl
Deteksi saat konten di situs web berubah dan dapatkan pemberitahuan melalui webhook atau email — tanpa perlu cron job, scraper, atau skrip diff. Gunakan skill ini setiap kali pengguna ingin melacak perubahan pada halaman, memantau harga pesaing, mendapat peringatan tentang lowongan kerja baru atau posting blog, memantau halaman dokumen/changelog/status, atau mengatakan "pantau", "awasi", "lacak", "beri tahu saya saat", "beri tahu saat X berubah", "kirim pesan jika", "kirim email saat", atau "kirim webhook saat". Sebuah hakim AI bawaan menyaring format, stempel waktu, dan...
officialweb-scrapingresearch
firecrawl-deep-research
firecrawl
Jalankan riset mendalam multi-sumber dengan Firecrawl. Gunakan saat pengguna meminta untuk meneliti suatu topik, membandingkan perspektif, menghasilkan briefing bersumber, menyelidiki pertanyaan teknis atau pasar, atau mensintesis bukti web dari banyak sumber.
officialresearchweb-scraping
firecrawl-research-papers
firecrawl
Temukan dan sintesis makalah penelitian, whitepaper, PDF, laporan teknis, serta sumber akademik dengan Firecrawl. Gunakan saat pengguna menginginkan tinjauan literatur, ringkasan makalah, lanskap penelitian, atau sintesis bersumber dari PDF dan publikasi ilmiah/industri.
officialresearchweb-scraping
firecrawl-market-research
firecrawl
Ekstrak metrik pasar, keuangan, pendapatan, industri, dan perusahaan dengan Firecrawl. Gunakan saat pengguna meminta riset pasar, tren industri, data perusahaan publik, perbandingan keuangan, riset pendapatan, atau laporan pasar terstruktur.
officialresearchweb-scraping
firecrawl-website-design-clone
firecrawl
Ekstrak sistem desain dari situs web mana pun menjadi DESIGN.md yang siap digunakan agen menggunakan bukti hasil scrape Firecrawl. Gunakan saat pengguna menginginkan warna, font, jarak, komponen, pola tata letak, atau panduan merek/antarmuka dari sebuah situs web sehingga agen AI dapat membuat situs web baru, meniru tampilan, atau membangun halaman yang terinspirasi dari desain tersebut.
officialdesignweb-scraping
firecrawl-knowledge-base
firecrawl
Bangun basis pengetahuan dari konten web dengan Firecrawl. Gunakan untuk dokumen referensi lokal, potongan data siap-RAG, dataset fine-tuning, cermin dokumentasi, korpora topik, atau markdown siap-LLM yang diorganisir dari sumber web.
officialweb-scrapingresearch
firecrawl-lead-research
firecrawl
Hasilkan ringkasan intelijen prospek pra-rapat dengan Firecrawl. Gunakan saat pengguna membutuhkan riset perusahaan, riset individu, berita terbaru, poin pembicaraan, titik kesulitan, atau persiapan penjangkauan sebelum panggilan penjualan, pertemuan kemitraan, percakapan dengan investor, atau wawancara pelanggan.
officialresearchweb-scraping