You're highlighting a real problem, but let me clarify what's actually happening:

**The actual risk:**
- Training data bias is real. If Reddit skews toward dramatic advice or breakup suggestions, models can absorb those patterns
- People *can* over-rely on AI for decisions that need human judgment
- An LLM confidently validating a breakup impulse could reinforce poor decision-making

**What's probably not happening:**
- LLMs don't mechanically output "Reddit's average take." Training data gets compressed into statistical patterns, not retrieved as direct recommendations
- A well-designed chatbot would refuse to directly advise breaking up, or would push back ("Have you considered couples counseling?")
- Most people using AI for relationship advice are *already* considering ending things, not being persuaded from scratch

**The real lesson:**
This isn't a failure of AI *per se*—it's a failure of **not treating AI like a search engine over your own judgment**. We shouldn't delegate high-stakes decisions to any single information source, algorithmic or not.

The solution isn't banning AI from relationship topics. It's:
- Understanding what AI is (pattern-matching tool, not wisdom oracle)
- Seeking diverse perspectives (AI + therapist + trusted people)
- Recognizing when you're outsourcing judgment you should own

What concerns you most here—the bias issue or the reliance issue?