2024-06-03

Activity Log

hackers.town: 2024-06-03 Mon 05:10

@ovid I’m doing more of a mix and match approach. For one thing, I haven’t clicked with ChatGPT, and I don’t have the budget for deep exploration of it. An existing Copilot subscription or Ollama are my primary options.

For another, my brain doesn’t work well with prompts yet. By the time I can articulate the question, that’s enough to get the shape of the solution in my head.

Finally, when I have reached that point, Copilot usually gives me suggestions that are at best mediocre and often just wrong. I’ve pondered that Copilot might expect some neurotypical communication style that’s just not natural for me.

But the code suggestions, weak though they usually are, can be enough to get me started on the sketch of a solution. I’ll write a throwaway sketch with assistance, then turn off Copilot and use ideas from the sketch to guide the real implementation.

hackers.town: 2024-06-03 Mon 06:52

Remembered that in a #Logseq style PKM graph, every node is a page. The details are in the edges. Updated code to reflect this in #KuzuDB.

hackers.town: 2024-06-03 Mon 06:54

@minego Good morning!

Or so I’ve been told.

hackers.town: 2024-06-03 Mon 06:58

@minego We’re still waiting for our furniture to reach us, and have been sleeping on an air mattress for several days.

The college dorm life experience is less fun at 50-something.

I mostly feel pain right now.

But it’s nice weather at least!

hackers.town: 2024-06-03 Mon 09:01

https://hackers.town/users/IamMrsGeek/statuses/112553514370259548

hackers.town: 2024-06-03 Mon 09:12

Now we’ve got namespaces! I’m seeing some funkiness in the graph explorer, with some pages not connecting to their namespace parents. But the query works fine in the REPL.

So I guess that means I’ve hit the point where Kuzu’s explorer won’t work for browsing the entire graph. Best to stick with the REPL except for very specific examples.

Something to keep in mind, since very soon comes adding every page’s blocks.

hackers.town: 2024-06-03 Mon 09:18

@minego Everything is still painfully in-process. But we arrived safely, and other than the cats’ dedicated avoidance of active ceiling fans, we’re all adjusting well to the general outline of our move.

But mover team is still in transit (arrival within the week?), the new place had not been cleaned effectively if at all, laundry piling up b/c hookups but no washer / dryer, old landlords giving grief about old place, budget stretched way past the limit, and of course my job search continues.

Oh and I’ve got about two weeks to get appointment under new insurance for ADHD meds.

Just a little overwhelming.

hackers.town: 2024-06-03 Mon 09:19

@Wrewdison I’d have to care enough to dig into developer view, but my hunch is they see a desktop browser, they assume a broadband connection.

hackers.town: 2024-06-03 Mon 09:20

@minego Oh I know. But we want things to calm down and get easier now.

hackers.town: 2024-06-03 Mon 10:25

@spinningthoughts so uh say if it was sufficiently like Neo4j – that one uses Cypher as the query language too, right? – how would I do a unique edge count?

Been wanting to learn graph database stuff for ages, but this is my first real dive into it.

EDIT: and though I haven’t seen the specific number of edge counts it can render, the docs repeat throughout that their Explorer view is best for prototyping and debugging, and has rendering limits.

hackers.town: 2024-06-03 Mon 10:29

@munin or hell even entered a dungeon room during a session managed by said DM.

hackers.town: 2024-06-03 Mon 10:33

@spinningthoughts I know I’d be making a different tool than Tana anyways – too heavily invested in local-first, for starters – but if I were to incorporate LLM tech in my dream PKM, it’d strictly be through Ollama, opt-in, and if you say “no thanks” during setup? It won’t keep trying to upsell.

hackers.town: 2024-06-03 Mon 13:36

Sorry I reported your email as spam. It was so grammatically forced and lacking in content that I assumed it was LLM-generated.

hackers.town: 2024-06-03 Mon 14:47

@philsplace I’m not super-thrilled about being in this picture either.

hackers.town: 2024-06-03 Mon 16:13

Leaning on the structures used when parsing my Logseq graph to just get the basic proof of concept.

Oh right. The concept I’m proving is that I can stuff content from the blocks in a page into Kuzu, then query for that content later.

hackers.town: 2024-06-03 Mon 17:58

And now there’s something like structure to the blocks in a page, with heading identification.

hackers.town: 2024-06-03 Mon 18:19

Always a little suspect about a company on LinkedIn when I see a near 100% overlap with what my shuttered former employer did, with labs in the same city, and zero network overlap.

hackers.town: 2024-06-03 Mon 18:19

Course, that could just be LinkedIn being stupid. Wouldn’t be the first time.

hackers.town: 2024-06-03 Mon 18:26

@jamie Yah, you’re one of the resident experts on the topic in my head.

I dabbled very lightly with Neo4j years ago, but the setup annoyed me enough that I never dug deep. Doing the embeddable thing with Kuzu is more my speed.

hackers.town: 2024-06-03 Mon 18:28

@jamie it needs a Ruby / Crystal layer, though.

Ruby + Polars + Kuzu + some-flavor-of-Glimmer would make for a load of fun.

hackers.town: 2024-06-03 Mon 18:37

@jamie That’s very much the idea. It’s intended to be complementary with SQLite and DuckDB (like SQLite with a tighter focus on OLAP).

Still a work in progress, but for my limited needs it’s doing great.

hackers.town: 2024-06-03 Mon 18:39

@jamie a Ruby GUI DSL basically. Implementations for lots of toolkits, with SWT being the most mature.

https://github.com/AndyObtiva/glimmer

hackers.town: 2024-06-03 Mon 18:41

@jamie DuckDB is still on my someday list. Most of the data analysis I’ve done is satisfied by one DataFrame-oriented library or another, or just SQL. And what those don’t satisfy, my direction has been more towards the graph stuff.

hackers.town: 2024-06-03 Mon 18:42

@jamie Glimmer is spiritually descended from _why’s Shoes project, so it feels very Ruby-like indeed.

hackers.town: 2024-06-03 Mon 19:52

@jamie I’ve only done a couple toy learning applications with it (via the SWT and LibUI implementations), but those were a lot of fun.

hackers.town: 2024-06-03 Mon 20:36

I use Git-backed sync for Logseq. Been writing a lot of code based on the notes I was writing a couple weeks ago. But my Github graph looks like I’m being less active, because my code has lower churn than my notes.

hackers.town: 2024-06-03 Mon 21:15

@spinningthoughts thanks! The Kuzu docs seem best-suited for someone who knows what they’re looking for. Now that I’ve learned enough about the very basics of Kuzu, I might visit GraphGists and do tweaked versions of those.

hackers.town: 2024-06-03 Mon 21:40

@spinningthoughts I’m also looking towards a data-assisted approach to note cleanup. I liked Cozo’s claims about what it was capable of, but their SQL contrast examples were so bad I got distrustful of the claims. Went with the more modest Kuzu, at least to start.

hackers.town: 2024-06-03 Mon 22:07

@spinningthoughts right in the tutorial first steps:

SELECT * FROM (VALUES(‘hello’, ‘world’, ‘sql’)) as unnecessary_alias;

To contrast with

?[] <- [['hello', 'world', 'Cozo!']]

I’m no data expert but I know the smell of BS

select ‘hello’, ‘world’, ‘sql’;