My confidence varies a lot by topic and by what you provide. I don’t have a single “overall confidence,” but here’s how it generally breaks down and how you can tell when to trust (or double‑check) me.

## When I’m usually **high confidence**
- **Well-established facts and concepts**: basic math, widely taught science, common programming patterns, standard definitions.
- **Deterministic tasks**: calculations (when clearly specified), code refactors, grammar/style edits, summarizing text you provide.
- **Tasks grounded in supplied material**: if you paste a document, I can quote/analyze it reliably.

## When I’m **medium confidence**
- **Open-ended judgments**: “best” tools/approaches, design tradeoffs, strategy advice, interpretations.
- **Areas with fast change** (tech, product details, policies), unless you provide current sources or context.

## When I’m **low confidence / needs verification**
- **Niche or highly specific factual claims** (e.g., “What did X say on date Y?”).
- **Anything requiring up-to-the-minute info** (news, prices, availability, current regulations) unless you provide sources.
- **Legal/medical/financial advice** beyond general information—should be checked with a qualified professional.
- **Exact quotes, citations, or “what’s in this paper”** if you don’t provide the text. I can produce plausible-sounding but incorrect details if forced to guess.

## Why errors happen
- I generate answers from patterns in training data; I don’t “know” facts the way a database does.
- If a prompt is ambiguous, I might assume details instead of asking (you can tell me to ask clarifying questions).
- I can “hallucinate” specifics (names, numbers, citations) when the truth isn’t strongly constrained.

## How to gauge confidence in a given answer
You can ask me to:
- **Provide uncertainty/assumptions**: “What assumptions are you making?”
- **Show reasoning or steps** (where appropriate): “Walk through how you got this.”
- **Give sources** (if you need verifiability): “Cite reliable sources or say you can’t.”
- **Offer alternatives**: “What are 2–3 plausible answers and what would distinguish them?”
- **Identify what would change the answer**: “What info would you need to be sure?”

If you tell me the domain (e.g., medicine, coding, history) and what you’re using the answer for (casual vs. high-stakes), I can calibrate how cautious I should be and how much verification to recommend.
