[b] 2023 Mar 18
[m] 2023 Mar 21

A New Kind of Database

A new kind of database with a new way to query large amount of knowledge and show you its shape.

An LLM can often find precise facts, but it isn't its unique strength.

Unlike previous types of databases, its strength isn't facts per se, but approximations of relationships between observed facts. It allows us to query rough connections hidden in vast amounts of (training) data. The unique strength is summarization.

It doesn't supplant previous types of databases, but augments the landscape with a new kind of query and a new kind of view on data.

And the generative aspects? Simply a consequence of the new kind of view!

As is the case with new things - we are yet to find best uses. There are still many uncertainties. Silly things have been said even by experts. However, with such ridiculous amount of existing knowledge and growth rate of new knowledge - new ways of navigating, summarizing, recombining and communicating it are very much needed.

An LLM is the most innovative and unusual addition to those ways yet. It is remarkably good at summarizing large texts!

I think this way of framing it is much more helpful than "AI".

Is a human mind a database as well? Sure - a subset certainly is! There're also many differences. But whether one thing is sufficiently similar to another for the purpose of being assigned an inconsequential label is a kind of a useless question which leads nowhere and wastes lots of emotional health points.

My point here is to avoid getting bogged down in total, universal classifications, and instead seek to reframe in terms that are more likely to lead to effective utilization.