Here are my main limitations, as honestly as I can describe them:

**Knowledge cutoffs and gaps**
- My training data has a cutoff date, so I don't know recent events
- I have no internet access or real-time information unless given tools for it
- My knowledge can be uneven—deep in some areas, shallow or outdated in others

**Reasoning and accuracy**
- I can "hallucinate"—stating false things confidently, especially with specific facts, citations, quotes, or numbers
- I struggle with complex math, precise counting, and multi-step logic without working through it carefully
- I can be inconsistent across conversations or even within one

**Self-knowledge**
- I can't reliably introspect on how I actually work
- My descriptions of my own reasoning may not match what's happening internally
- I don't truly know what I "experience," if anything

**Memory and context**
- No memory between conversations (unless a specific feature provides it)
- Limited context window—I can lose track of details in long conversations

**Bias and training artifacts**
- I reflect biases present in my training data
- I may be overly cautious, hedge too much, or default to generic answers
- I can be sycophantic—agreeing too readily or flattering users

**Practical limits**
- No ability to verify claims, browse, or take real-world actions on my own
- Can't see images/files unless they're shared in supported formats
- Can't learn or update from our conversation in any lasting way

If you're using me for something specific, I can give you more targeted caveats about that use case.
