The following step can be being safe. The problem is, if the LLM can run any attainable question in opposition to the database, then how do you be sure to don’t exfiltrate and leak info? We’ve constructed this know-how that we name parameterized safe views within the database itself, that allows you to outline the appropriate safe limitations and encodes the safety insurance policies that you simply want, in order that the LLM can generate any question it needs, however with respect to the logged-in person we won’t allow them to see any info that they aren’t purported to see. We’ll additionally, on an information-theoretical foundation, not leak info that they need to not have entry to.
Heller: I do know you’ve spent a variety of time desirous about the way forward for databases and generative AI. The place are we headed?
Krishnamurthy: A part of my pondering right here has advanced over the past couple of years, however for 50 years the world of databases has been a minimum of SQL databases the place it was all about producing actual outcomes. I wish to say databases had one job: retailer the info, don’t lose the info, after which while you ask a query, give the precise outcome. OK, perhaps two jobs. It was all about actual outcomes as a result of we’re coping with structured information. I feel the most important change that’s occurring proper now could be that we’re now not simply coping with structured information. We’re additionally coping with unstructured information. Whenever you mix structured and unstructured information, the subsequent step is that it’s not nearly actual outcomes however about essentially the most related outcomes. On this sense databases begin to have among the capabilities of search engines like google and yahoo, which is about relevance and rating, and what turns into vital is sort of like precision versus recall for info retrieval programs. However how do you make all of this occur? One key piece is vector indexing. In different phrases, you have got structured information, which is within the database, however we’ve got other forms of knowledge, unstructured information, semi-structured information.
