November 2024

I really like boring technology. Software that has been tested for decades, does what it is supposed to robustly and fast, and is known by every person and machine on the planet. Technology like relational databases and SQL, HTTP, HTML, a well-known programming language. Throw LLMs into that mix, and you've got a potent cocktail!

It’s no wonder LLMs are taking the world by storm: they feel like magic, can do incredibly useful things that we could only have dreamed of (checks calendar) about two years ago, and they’re available to anyone that can write somewhat coherent sentences.

And while we developers are trying to wrangle LLMs into our mental model of how computers work, myself included, basically everyone can now call themselves a prompt engineer and maybe even build digital products that way.

Do you want to know how I keep up with the rapidly changing world of LLMs while also producing useful software? Okay, you already know from the title, but let me spell it out for you here: I don’t engage in a lot of other fancy software.

“Go is my hammer, and everything is a nail”, and I love building stuff on top of SQLite on a beefy but conceptually simple VPS (there’s an old acronym!) from Hetzner. Yes, I use cloud technologies when I need to (nothing beats S3/R2 for storing blobs of data, in my opinion), but I couldn’t tell you the whole suite of products available on AWS even if I wanted to.

(Oh yeah, Postgres is also great if your data needs are beyond CRUD and the usual primitive types.)

Now I can spend my time learning about building LLMs into applications, including but not limited to prompt engineering, evals, use of embeddings, multi-modality of LLMs, and the list goes on… What an exciting time to be alive! 😁

As a parting gift because you’ve come this far, my current list of interesting information on language models: