Til 2025-03-31

There is a visualization source for learning what is behind the LLM, and I learned that I don’t understand anything about it. It’s fundamentally new to me.

We can change font in Slack.

There is a product named ParadeDB which brings the power of Elastic Search into Postgres. It has an extension for postgres named pg_search. The interesting part here is searching pg_search only results in a gem for Rails. Even Claude doesn’t know this.

ParadeDB is NOT a single Postgres extension, so I can just use it by a simple SQL statement like

  CREATE EXTENSION IF NOT EXISTS  ....

In fact, it contains a set of custom extensions to make it work. So, with my current GCP Cloud SQL, it’s not a straightforward setup.

Neon db now supports this, that was good news, hopefully,y Cloud SQL can support this soon.

Now I can truly understand what context window is in terms of LLM, thanks to the explanation from Gemini. The surprising thing is it’s what I understood from the beginning from the first time I saw this word. So just a confirmation of my understanding. 1M tokens context window from Gemini 2.5 is awesome.

Neon ability to restore the database is insane, see this blog post. The approach is they use the “branch” model rather than the traditional restore operation. The Neon’s branch approach works most similarly to the Git branch! So you checkout a new branch, you get the data on that branch immediately, at the high level, it explains how fast it is. This is a brilliant approach. What I don’t understand is WHY Neon first commit was in 2021. I thought about something like this when I was a junior developer from many years ago, crazy.

Fixed the re-render problem of videojs, that is annoying. Videojs has support options for hotkey, good to know.