Comment by bloopernova | Hacker News Reader

Comment by bloopernova | Hacker News Reader

I feel like there's a lot of knowledge or information that we're "leaving on the plate". For instance, the sites we visit, the files we edit, the branches and PRs we create, etc etc. All of that is related, but it feels like that context is being lost or discarded.

An example might be: I have to include new AWS resources in a deployment, so I look up information about them, find examples and read about potential problems, security information, etc etc. That then becomes edits in a terraform file somewhere, with a Jita ticket, my own knowledge database (Emacs org-roam files in my case, Obsidian etc for other people). Then the feature branch gets a PR to dev, we might discuss changes in Teams (ugh) or a meeting. All of that seems ripe to be linked together conceptually, but the computer has no way to do that.

It makes me wonder if that could be fed into the right machine learning thing to at least start tracking this sort of work stuff. Heck just synchronizing my Firefox bookmarks (ff lets you tag your bookmarks) with my org-roam instance's tags would be useful. Tagged files in my knowledge base could be automatically linked to similarly tagged bookmarks.

I like these pieces of my digital footprint to not be connected. There is no need to track everything.

Do you not want them connected, or do you not want the connections shared and potentially used against you?

I like them not to be collected or connected. I don't trust those collecting such data.

I typically write up all of that in my documentation somewhere. Stuff like "first thoughts are this approach might work, talked to person who had this idea, looked at this link and found this info, decided to go with this approach because of factors x, y, z". This isn't the primary user-facing documentation but a subpage or something that's helpful a couple of years down the line

It's like a book titled "A History of [Object]" that traces what solved problems before the object, issues with old solutions, the emotional, financial, etc state of the inventor, why they chose this solution over that one, how the object was adopted and improved afterwards, other inventions spawned off the object, etc. Capturing the history of the object requires capturing the context around the object too

My thoughts on this are to slow down and document and explore that knowledge and information. If it is really valuable, the "loss" in efficiency from slowing down will be offset by the gain in skill/utility from really grokking the stuff.

If it's not...then there's really nothing "left" on the table — if ever turns out to be valuable, you'll probably come across it again, when needed.

I constantly get a similar feeling. I'm speeding around from task to task, just grasping enough to get the current task done so I can get to the next one and the next one...

And somehow this is value-creating? Apparently it is, but it seems almost accidental, at that rate.

I'd rather slow down and appreciate the value as it moves through me, into whatever I'm doing.

I usually get more from the process, at the same time.

It's like...if "less is more," then "more is less."

Reminds me of a floating point number. The bigger or smaller they get, the less accurate they become.

If you're chunking on a ton of data and tasks, you're getting less out of it. At a certain point, none of it even seems to enter your brain at all.

Basically, this is what college should be teaching you - how to research. What good does are useless facts? I don't want to walk around cluttered with a dictionary - I want to know where to look in that dictionary. Obviously in the sciences there are facts that you should know, but even with math, its more about how to derive the formula, than actually memorizing it. I mean, their called "Research Papers" right?

Totally agree. I remember the phrase “learning how to think” being thrown around.

I also remember not being explicitly taught that.

It sort of seems like trying to find enlightenment by chopping wood and carrying water at a monastery.

If critical thinking is something that spontaneously emerges in a learning environment, maybe we shouldn’t sell it as a benefit. “Some students experience deep insight into the nature of the mind. Results not typical.”

I've been thinking of something like this since LLMs became popular. I've toyed around with some proof of concepts, but haven't had the time or motivation to work on it lately. I love the idea of tagging everything and showing connections when you're searching for things. Also semantic search would be great, like "blue website with information about databases I read last week" would be super powerful in my opinion.

I really love the idea of digital knowledge bases, but as you said, I think we're leaving a lot on the table. I need to get back to my project of a user-owned-data knowledge base.

What kind of approach did you take? I was thinking along the lines of requiring something like rewind.ai or some program that autoscreenshots your screen at a set interval (or originally a recorded video split into several images later) and having a vision-capable model (particularly specialized in UIs) describe these set of images in order to build a dataset of images-tags-description and the like.

There's also libraries like trafilatura in Python featured here in HN some time ago that could extract content from websites to help augment the data.

I've had similar thoughts but over time you'd just end up with a private copy of the internet. You'll still have to search for the information anyway, so I'm not sure what the benefit is. Searching your knowledge base for "the thing I did yesterday" vs "how to sync Azure to AD" seems basically equivalent to me. You're just creating yet another thing to search.

That's a good point, you'd absolutely want to get away from adding another burden to the human.

Seeing relevant bookmarks when I'm viewing a specific note in my database could be useful though. And finding pull requests related to a subject might also be useful.

So the idea would be to reduce the number of searches performed by the human. Automate and enhance rather than dump and forget.

Yeah, but your private copy would be more like "The internet: The Good Parts" (assuming you had a way to not store what you immediately dismissed as garbage; maybe only include pages with a dwell time of 15-30s or more.) That's enormously valuable (and why I've implemented it before - but in conkeror, which didn't survive the death of xulrunner - so now I use pinboard and text files and logseq, which are pretty good but a lot more work.)

To whatever extent something like this can be done locally, I'd probably pay a monthly sub for it if its good enough. But I wouldn't want any of that leaving my machine, we get tracked and profiled enough as-is imo.

Yeah, this is worth at least as much as Kagi or Copilot is to me right now.

Working on something like that, but there’s still a good amount of work to do

Reply on news.ycombinator.com