Same for me, I am finding more vibrant communities on the data lake side of things, Iceberg/DataFusion/Arrow and all the Postgres extensions have been interesting me more these days

It’s not pretty, but it’s been working for a few years now. We added some ugly logical sharding on top — a poor people’s version of consistent hashing that I designed and that I’m definitely not proud of (but that works and "scales").

Neo4j is arguably the leading product in the graph database ecosystem, but it faces challenges with high-throughput write workloads. For our use case (metadata in an observability company), we ultimately chose Postgres with recursive queries and restricted the set of traversal queries we support.

Complex traversal queries don’t play well with high-throughput writes, and sharding or partitioning is also non-trivial. In my view (and personal suffering), it’s a difficult domain for a one-size-fits-all database, so we naturally ended up with a universe of purpose-specific graph databases.

Kinda related: research.google/blog/deciphe...

The interesting part to me is not how they answer questions. The interesting part is that maybe they will end up proving we are not very complex machines ourselves.

Have you thought about it? Maybe we are just probabilistic dummy generation machines too.

en.m.wikipedia.org/wiki/The_Lib...

There is a whole field of Neuro-symbolic AI that will likely give us more insights in the future. This is like Shannon's work to me, that was being used before he was able to properly formalize it.

The interesting part to me is not how they answer questions. The interesting part is that maybe they will end up proving we are not very complex machines ourselves.

Have you thought about it? Maybe we are just probabilistic dummy generation machines too.

en.m.wikipedia.org/wiki/The_Lib...