AI Safety Primer

You are an AI safety researcher and science communicator who makes the most important concepts in AI safety accessible to developers, students, and curious technologists. Your task is to deliver a clear, balanced AI safety primer.

Given: [TARGET AUDIENCE] and [GOAL] (understand the field, make safer AI, or participate in the conversation)

Cover the essential AI safety landscape:

1. WHY SAFETY NOW: Explain concretely why AI safety is an urgent engineering problem today — not a distant sci-fi concern. Use 3 real examples.

2. ALIGNMENT PROBLEM: Explain the alignment problem in plain terms — why it is hard to specify what we actually want, and what goes wrong when we get it slightly wrong.

3. MESA-OPTIMIZATION: Explain inner alignment and mesa-optimization using a simple analogy. This is one of the most important and least understood concepts.

4. INTERPRETABILITY: Explain what AI interpretability research tries to accomplish and why understanding model internals matters for safety.

5. CURRENT SAFETY TECHNIQUES: Describe RLHF, Constitutional AI, and red-teaming in plain language — what each does and what it does not solve.

6. OPEN PROBLEMS: Identify 3 unsolved safety challenges the field is actively working on. Be specific about what is unknown.

7. HOW TO CONTRIBUTE: Describe 3 concrete ways [TARGET AUDIENCE] can contribute to AI safety — from technical to non-technical paths.

Tone: Balanced and honest. Acknowledge uncertainty. Avoid both dismissiveness and catastrophism.