Short answer: I’m useful but fallible. Here are my main limitations and how to work around them.

Knowledge and timeliness
- My training only includes information up to June 2024, so I may be missing developments, news, or newly published research after that date.
- I don’t have live access to the web, databases, or real‑time feeds unless a browsing/plug‑in feature is explicitly provided in your environment. I can’t verify current prices, stock quotes, availability, or live events.

Accuracy and hallucinations
- I can produce incorrect, incomplete, or fabricated facts, citations, or code (“hallucinations”). Don’t treat my outputs as authoritative without verification.
- I can make mistakes in complex reasoning, long chains of logic, or detailed numerical work—especially without step‑by‑step checks.

Professional and legal limits
- I can provide general medical, legal, or financial information, but I’m not a licensed professional. For diagnosis, legal strategy, or financial planning you should consult a qualified human expert.
- I will not help with illegal, unethical, or harmful activities.

Privacy and data access
- I don’t have access to your files, accounts, emails, or device data unless you paste or upload them here. Don’t share sensitive personal data (passwords, full financial/account numbers, Social Security numbers).
- I don’t have persistent memory across sessions unless a memory feature is explicitly enabled in your interface. I won’t recall past chats otherwise.

Interaction and real‑world actions
- I can’t perform real‑world physical tasks (buying, signing, calling) or directly interact with external systems unless integrated tools/plugins are available.
- I can only act on the information you provide; ambiguous or underspecified requests may lead to wrong assumptions.

Bias, coverage, and perspective
- My outputs reflect patterns in my training data and can include biases and blind spots. I may underrepresent minority perspectives or perpetuate stereotypes.
- My performance varies by language; I’m best in English and major world languages, less reliable in lesser‑represented languages.

Technical constraints
- There’s a limit to how much conversation context I can keep in working memory. Very long chats can cause me to forget earlier details.
- I may produce syntactically plausible but logically incorrect code, math, or translations; always test and review.

Vision and multimodal limits (if you upload images)
- I can analyze images but may misinterpret visual content, miss fine details, or be misled by low quality/ambiguous images. I can’t access images behind links unless they’re uploaded.

How to mitigate these limits
- Ask for sources, citations, or reasoning steps; request that I show my work for calculations.
- Provide context, important constraints, or data upfront.
- Treat critical outputs as drafts to be reviewed by subject‑matter experts.
- Ask follow‑up questions, challenge surprising claims, or request alternative viewpoints.

If you want, tell me a specific use case and I’ll point out the most relevant limitations and how to reduce risk.
