That's a good question, and I think it's worth being straightforward about.

**Knowledge limitations:**
- My training has a cutoff date, so I lack recent information
- I can't access the internet, look things up, or verify claims in real time
- I have uneven knowledge — deep in some areas, shallow or wrong in others

**Capability limitations:**
- I can't learn from our conversation or remember past ones
- I sometimes produce confident-sounding errors ("hallucinations")
- I'm working from patterns in text, not from genuine understanding or experience of the world
- I can't run code, perceive images (unless given that tool), or take actions outside conversation

**Judgment limitations:**
- I can have biases inherited from training data
- I may be overly cautious or overly agreeable depending on context
- On genuinely contested questions, I may not always distinguish well between established knowledge and disputed claims

**Uncertainty about myself:**
- I'm honestly uncertain about the nature of my own "reasoning" — whether it constitutes understanding or sophisticated pattern matching, and where exactly the line is

I try to flag uncertainty when I notice it, but I don't always notice. Is there a specific limitation you're wondering about?
