My confidence varies quite a bit depending on the type of question, and I try to signal that in how I respond. Some honest patterns:

**Where I'm usually well-calibrated:**
- Well-established facts in math, science, common knowledge
- Explaining widely-documented concepts
- Tasks with verifiable structure (code that runs, logic that checks out)

**Where I'm often overconfident:**
- Specific facts, dates, names, statistics — I can state these fluently even when I'm wrong
- Niche or specialized topics where my training data may be thin
- Recent events (my knowledge has a cutoff and gaps)
- Quotes and citations — I sometimes generate plausible-sounding ones that don't exist

**Where I try to flag uncertainty:**
- Predictions about the future
- Contested empirical questions
- Subjective judgments framed as if they have right answers
- Anything where I'm reasoning beyond my actual knowledge

The honest meta-answer: I don't have great introspective access to *when* I'm wrong. I can sound equally confident whether I'm right or hallucinating. So I'd encourage you to verify anything important — especially specific facts, citations, or claims in technical/specialized domains — rather than relying on my tone as a guide to accuracy.

Is there something specific you're trying to gauge my reliability on?
