Ask what should change when the equation is manipulated, then let the visualization test that expectation.
Foundation Lab
Deliberative Alignment
Trains models on explicit specifications rather than implicit reward shaping
#85DeliberativeScaling & Alignment
key equation
\max_\pi \mathbb{E}[r_{\text{help}}] + \lambda \mathbb{E}[v_S]Selected Foundation Object
Keep the equation fixed; move through the evidence.
Concept 85 of 100DeliberativeScaling & Alignment / Phase 12: Advanced alignment & safety research
Current question
Model retrieves relevant policy text, reasons about it, then responds
\max_\pi \mathbb{E}[r_{\text{help}}] + \lambda \mathbb{E}[v_S]Use the runnable panel, the key equation, and canonical papers as separate forms of evidence for the same object.
The useful learning product is the reusable mechanism you can carry into another model, paper, or engineering tradeoff.
Use prerequisites, dependents, and semantic links to repair the next gap without leaving the object behind.
Why It Matters for Modern Models
- Trains models on explicit specifications rather than implicit reward shaping
- Enables auditability: which policy clauses were consulted?
- Reduces over-refusal while improving jailbreak robustness
What Tutorials Skip
What is still poorly explained in textbooks and papers:
- Model retrieves relevant policy text, reasons about it, then responds
- Like Constitutional AI but with explicit spec document in context
- Pareto frontier: helpfulness vs safety vs over-refusal
Interactive Visualization
Core Math (Optional Deep Dive)
If you want intuition first, start with the key equation and the visualization. Come back here for the full walkthrough.
Key Equation
Train model to reason over safety specifications :
Constrained optimization view:
Lagrangian form:
where scores compliance with spec text .