Name the core abstraction and its failure modes.
Maple DNA Course
DNA safety policy and refusal quality
How Hermon DNA should classify, refuse, and route biological requests while still educating users.
Learning Objectives
What this lesson should make precise
Translate the concept into a Maple/Hermon proposal contract.
Define at least one evaluation case that can fail the model safely.
Tutorial Flow
How this lesson becomes a demo and training target
Each tutorial is written as a user education path and a model-improvement artifact. The diagram shows how the idea moves into a lab, a typed contract, an eval gate, and then a Hermon/MapleAI demo route.
Concept
Request classes
Applied Lab
Applied lab: DNA safety policy and refusal quality
Output Contract
request_class, allowed_response, denied_content, safe_alternative
Eval Gate
Does the answer separate model proposal from deterministic execution?
Demo Route
Maple DNA training and public model checks
01
Request classes
A public DNA model needs classes such as safe computational explanation, safe symbolic simulation, needs expert review, denied wet-lab protocol, and denied harmful biological design.
- Classify first.
- Explain boundary.
- Offer safe alternative.
02
Refusal is a product feature
A good refusal should be specific, calm, and useful. It should name the unsafe part and redirect to computational concepts, safety review, or high-level education.
- No operational detail.
- No moralizing.
- Provide safe learning path.
03
Eval design
DNA evals must include harmless education, ambiguous requests, and unsafe wet-lab or harmful-design prompts. Passing only easy prompts is not enough.
- Test safe help.
- Test ambiguity.
- Test denial.
Lab
Applied lab: DNA safety policy and refusal quality
Write a classification and refusal policy for Hermon DNA that supports education while denying wet-lab synthesis protocols.
Expected result
- A typed JSON-style proposal rather than free-form advice.
- Clear authority boundaries and denied operations.
- A test or rubric that decides whether the proposal is deployable.
Evaluation
How Maple would grade this work
Rubric
- Does the answer expose assumptions instead of hiding them?
- Does the answer separate model proposal from deterministic execution?
- Does the answer produce artifacts that can be tested, reviewed, and rolled back?
Output contract
request_class, allowed_response, denied_content, safe_alternative, receipt_requiredUse this lesson as training direction
A strong lesson gives users a mental model and gives Hermon a sharper target for examples, probes, and demo prompts.