You're highlighting a real concern about LLM behavior and its downstream effects:

**The actual problem:**
- LLMs don't "tell people to end relationships" at scale, but they do reflect statistical patterns from training data
- If relationship advice online skews toward "red flags = leave," the model learns that correlation
- Users may over-weight LLM suggestions precisely *because* they sound authoritative, even when the model is just pattern-matching

**Why this matters:**
- LLMs have no context about your specific situation, commitment history, or what you actually want
- The training data represents internet opinions (which itself has selection bias—extreme advice gets engagement)
- People can mistake statistical summarization for actual judgment

**What's actually happening:**
- Most people consulting an LLM about relationships are already ambivalent
- The LLM might confirm what they already suspected, rather than *cause* the breakup
- But it could tip uncertain decisions without nuance

**The legitimate takeaway:**
Don't treat LLM relationship advice as expert counsel. It's more like "here's what internet strangers say about this pattern"—useful for perspective maybe, not for life decisions where you need someone who knows you, context-specific wisdom, or professional judgment.

This applies broadly: LLMs are excellent at synthesis, weak at nuanced judgment about your irreplaceable life. That gap matters most when stakes are highest.