AI and LLMs · Middle · Question 61 of 75
Why are vector or embedding weaknesses listed as an LLM risk?
- 1Leaky or poisoned vector stores hurt answers and privacy
- 2Vectors make models answer only in numbers from then on
- 3Embedding models can't run on servers
Show the answer
Correct answer: Leaky or poisoned vector stores hurt answers and privacy
Explanation
Issues include missing access control, cross-tenant leaks and inversion attacks that recover text from embeddings.