80 percent accuracy on BIRD is impressive, but text-to-SQL is always a huge privacy risk. When an AI generates queries, how is it handling the underlying database schemas or sensi…
80 percent accuracy on BIRD is impressive, but text-to-SQL is always a huge privacy risk. When an AI generates queries, how is it handling the underlying database schemas or sensitive PII? I'd want to know if these models are being trained on enterprise data or if they’re strictly sandboxed.
Tbh, unless there’s robust encryption and strict data minimization, letting a model touch production databases feels like a massive security headache waiting to happen. Does Google have a clear audit trail for these queries?