Lesson 3: Trust and AI
Risks and hallucinations
Hallucinations can be funny and harmless. But what are some situations where hallucinations could be risky?
Risky uses of LLMs
- When it is used for decision making or when asking advice, like asking an LLM if a mushroom is safe to eat
- When we are given false information that gets shared with others, like a lie about someone being shared
Hallucinations don't just happen in LLMs with Text responses. Hallucinations happen in other modes too: Video, Audio, Images and Code.
Sometimes hallucinations are easier to notice in those modes than they are in Text.