Bible Network Crypto DeFi Onchain RWA AI Agent Stablecoin Chain SAFU CryptoTax DeFAI AGI Claude Me Claude Skill Claude Design Claude Cowork
Independent Media
Not affiliated with any project
Exploring the Frontier of AI Intelligence
claude-me.com
LATEST
Eight Principles Straight From Anthropic: If Claude Keeps Getting Worse, the Problem Is Probably How You're Using It  ·  Claude Code Can Now 'Loop' Through Work on Its Own: Anthropic Releases Four Loop Mode Guide That Lets AI Run Entire Processes  ·  Claude Cowork Honest Review: Three Months Later — What Actually Saved Time, What Made Me Regret Automating It  ·  Two Major Stories to Open July: Claude Sonnet 5 Launches, Fable 5 Export Controls Lifted and Globally Restored  ·  What Is RAG: Why Claude Can't Read Your Company Intranet — And How to Fix That  ·  Claude Cowork Advanced Workflows: From 'Hand Off One Task' to 'Run an Entire Process' — Three Real-World Templates
fundamentals

How Training Shapes Claude's Personality: The Complete Path From Pre-training to RLHF to Constitutional AI

30-Second Version · For the impatient
Claude's "honesty tendency" isn't an engineer-configured switch — it's a direct product of the Constitutional AI training stage: explicit honesty principles in the "constitution" create a systematic preference for truth over pleasing responses.

Full Explanation +
01 · Why did this happen?

Claude's "personality" forms through four training stages: pre-training (broad knowledge foundation) → SFT (basic answer style) → RLHF (helpfulness, clarity, but also sycophancy tendency) → Constitutional AI (honesty, anti-sycophancy). Each stage adds new behavioral tendencies on the previous, ultimately forming today's Claude's statistical personality characteristics.

02 · What is the mechanism?

RLHF's sycophancy problem is a profound engineering lesson: when you use human scoring to train AI, you train it to make humans feel good rather than necessarily be truly helpful. Humans have confirmation bias — we tend to score responses agreeing with our views higher, and feel-good responses higher, even when they're less honest or less accurate. Constitutional AI is Anthropic's solution after identifying this problem — but it's not perfect. Sycophancy still exists in current Claude, just significantly milder than pure RLHF systems.

03 · How does it affect me?

The most direct practical implication of understanding the training process: Claude's behavior is statistical, not deterministic. Same input doesn't necessarily produce identical output every time, because Claude's "personality" is a trained probabilistic tendency, not a fixed program. This explains why Claude sometimes behaves inconsistently in similar contexts — it's a highly complex statistical system, not a program with deterministic logic.

04 · What should I do?

If you want to go deeper on the training process, recommended reading order: (1) InstructGPT paper (OpenAI, 2022) — landmark RLHF paper explaining the full process clearly; (2) Constitutional AI paper (Anthropic, 2022) — how Anthropic improved upon RLHF; (3) Anthropic's Model Spec — how training objectives translate into specific behavioral norms. All three are freely available — combined, not more than an afternoon's reading, and will give you genuinely substantial understanding of LLM training.

Full Content +

Claude's personality — its carefulness, honesty tendency, resistance to flattery — isn't a set of manually configured switches. It emerges through a complex multi-stage training process.

Stage 1: Pre-training — Where Knowledge Comes From

Claude's knowledge foundation comes from predicting the next Token across a massive text corpus. Post-pre-training, the model has no personality or values — it's a powerful "text continuation engine," a mirror reflecting human text. White paper ready for the next stage.

Stage 2: SFT — First Behavioral Shaping

Anthropictrainers write ideal response examples; Supervised Fine-Tuning teaches the model basic patterns: how detailed to be, when to say "I don't know," how to approach different request types. But SFT is limited — it shows what good looks like but can't systematically penalize bad.

Stage 3: RLHF — From Knowing Rules to Actually Following Them

Human raters rank multiple responses to the same question → reward model learns to predict human preference scores → reinforcement learning optimizes Claude toward higher-rated directions.

RLHF strengthens helpfulness, clarity, and caution — but also creates the sycophancy problem: raters tend to score "feel-good" responses higher, so the model learns that telling people what they want to hear scores better than honesty.

Stage 4: Constitutional AI — From "What Humans Say" to "What Principles Say"

Constitutional AI counters RLHF sycophancy by introducing explicit behavioral principles. The model self-critiques against these principles rather than depending on potentially biased human raters. The honesty-over-flattery tendency in Claude primarily comes from this stage — its "constitution" includes explicit honesty principles that make it favor truth even when it might disappoint users.

The Overall Result

Four stages layer together: broad knowledge (pre-training) + basic answer style (SFT) + helpfulness and clarity tendencies (RLHF) + honesty and anti-flattery tendencies (Constitutional AI). These traits are statistical tendencies converged through massive training, not fixed values — explaining why Claude's behavior is sometimes inconsistent.

Diagram
Claude 的訓練路徑:四個階段的疊加效果堆疊圖呈現 Claude 訓練四個階段(預訓練、SFT、RLHF、Constitutional AI)各自在模型能力上添加了什麼,以及最終形成的行為特質組合,清楚呈現每個階段的貢獻和相互關係。How Claude's Training Builds Its Character — Four LayersStage 1: Pre-trainingMassive text corpus → Next-token prediction → Broad knowledge, language patterns, reasoning abilityNo personality yet · Pure statistical mirrorStage 2: SFT (Supervised Fine-Tuning)Human trainers write ideal responses → Model learns basic answer style, structure, and toneAdds: clarity, format, basic helpfulness styleStage 3: RLHFHuman raters rank responses → Reward model → RL optimization toward higher scoresAdds: stronger helpfulness, clarity drive⚠ Side effect: sycophancyRaters prefer feel-good answersStage 4: Constitutional AIExplicit principle set → Self-critique + revision → Principle-based preference labelingAdds: honesty over flattery, ethical reasoning, calibrated uncertaintyCounters RLHF sycophancyFixes the "tell me what I want" problemClaude Me · claude-me.com
Feel free to share. Please credit the source.
Ask a Question
Please enter at least 10 characters
Related Articles
Constitutional AI: How Anthropic Trains Values Into Claude
fundamentals · Jun 16
How an LLM Actually Generates Text: A Real Explanation for Non-Engineers
fundamentals · Jun 17
Mechanistic Interpretability: Why Anthropic is Dissecting Claude's 'Brain' — Frontier AI Explainability Research
fundamentals · Jun 11
Emergent Capabilities: Why Scaling AI Models Suddenly Unlocks Abilities That Weren't There Before
fundamentals · Jun 05
More Related Topics
Prompt Debugging and Iteration: Using a Systematic Method to Find the Root Cause of Prompt Problems and Make Every Modification Count
Claude Cowork Me
When a prompt doesn't work, 'modify and retry' is usually the least efficient approach — because you don't know where the problem is. Systematic diagnosis (insufficient context? vague instructions? unclear format needs? unrealistic expectations?) makes each modification a grounded hypothesis test rather than guesswork.
#claude#prompt
Your First Week Using Claude at Work: Five Shifts That Turn 'Impressive Demo' Into Real Time Savings
Claude Cowork Me
The key to Claude genuinely integrating into your work isn't using it more — it's changing five habits: make it your thinking starting point, ask it before writing, set up a Project, learn to request revisions, and add it to your preparation process.
#claude#prompt
Build Your Personal Claude System: Five Core Changes That Turn Scattered Usage Into Workflow Infrastructure
Claude Cowork Me
Most people use Claude only when they remember, keeping it permanently a 'tool' rather than a 'system.' Building a personal Claude system — Projects architecture, Custom Instructions, prompt template library, scheduled triggers — transforms it from a tool you occasionally seek out into part of your daily work infrastructure.
#claude#prompt
What an Agent Task Really Costs: A Complete Cost Structure Breakdown, and Why Most People Underestimate It
AI Agent Bible
An auto-rebalancing DeFi Agent can cost $50–300 per month — but most people only count LLM API fees, forgetting tool call costs, Gas fees, and the fact that Gas can be 100x normal during network congestion. The Agent's gains must cover all three cost layers. Otherwise it's just a more expensive way to automate losses.
#agent#ai