LOGBOOK

HELP

Quiz Entry - updated: 2026.07.14

What are the main benefits and limitations of differential privacy?

Benefits: provable, background-knowledge-proof protection with accurate group insights. Limitations: reduced precision, compute overhead, and a query/privacy budget that depletes.

Benefits Limitations
Mathematical guarantees (provable, not just promised) Reduced precision for individuals/small subgroups
True privacy — neutralizes linkage attacks Computational overhead to add/compensate noise
Accurate group insights preserved Design trade-offs — more privacy = more noise, less granularity
Builds user trust (participate without fear) Limited queries — running many on the same data erodes the guarantee

The last limitation is the killer for interactive systems: every query spends the privacy budget, so DP suits well-defined, repetitive queries where the budget is managed, far less so open-ended exploration.

Tip: DP is great for "how many users did X?" asked a bounded number of times; it struggles with "let me poke at this dataset however I like."

From Quiz: PRIVACY / Privacy in AI & ML — Differential Privacy, Synthetic Data & LLM Security | Updated: Jul 14, 2026