I decided to try the AI train and develop something without actually writing any line of code.
The result was very satisfying, https://rss.sabino.me - a rss new aggregator, with a summary to save some time on news that I do not want to read, but am mildly interested.
I will not say the same about the code quality tough
I really like the idea behind the brutalist report [1], mostly related to their claim of "headlines delivered without bullshit".
But I wanted even less interaction with most websites, as most of the headlines are related to something I am just mildly interested on, to the point of wanting to know more, but not interested enough to the point I feel like committing to reading the whole thing before knowing a bit more what it is about.
Anyway, I build this thing that parses the RSS feeds and do all summaries locally through ollama or whatever other OpenAI API compatible llm platform you have (e.g. LM Studio), and auto updates automatically twice a day through GitHub action scheduler
Interesting that two complete opposite views are in the first page today. The other one being "AI isn't going to kill the software industry" https://news.ycombinator.com/item?id=42810175
I wonder if ideas from the embedded steno community [1] [2] could not be used, as this type of chording/key stroke replacement is the basis of how it works.
Temporal graph visualization is quite lacking. It probably depends what one time slice of your network looks like.
If you have 10 to 20 nodes without a ton of edges you could use a fixed circular layout for the nodes on a timeline that just the edges change over time.
More than this and you start getting into hairball territory even without the changes over time. A hairball changing over time is even more useless than a hairball on its own.
The standard force directed layout is really quite useless other than seeing the global structure of the graph. The uselessness is more obvious when you try to change it over time. I suspect the layout is so standard because the visualization looks so cool.
Most data visualization though has this same problem. There is almost this property that the cooler and more beautiful the visualization is, the more useless it is as far as containing any insight about the data itself.
Picture musical notation, 5 horizontal lines where the y-axis is the note (higher on the staff, higher note) and the x-axis is position in time.
Say you have 5 event types, which occur with various frequencies over time. Plot them as dots on the staff.
Draw lines between events which link to each other.
If you draw the lines for the staff, they should be faint, only to help classify the event.
And you can of course have more than 5 event types, just add more lines. Hopefully less than 20 lines, otherwise this starts to get visually very messy.
For print/static graphic? An Ishikawa diagram paired with a, probably keyed, network diagrams(s). For video or interactive there would be more flexibility. Your delivery method ultimately makes many choices for you.
And I strongly suggest that you contact Kay Xu <Kai.Xu@nottingham.ac.uk>, who is doing research on sensemaking [1] [2] and berrypicking [3], I think he is currently working on newer and better version of his approach with browser extensions (as opposed to a separate renderer), and you both would benefit from collaboration.
As I mentioned, it is a research project, so I would not expect production ready code or multi-browser support. They are indeed not supporting the old version, as they are rewriting almost everything from scratch.
Last time I had contact with them, they were exploring using Plasmo [1] as building block for the extension, instead of doing everything vanilla as they did in the 1st version, which would offer cross-browser support out of the box.
But meanwhile, you can check the code [2] and add the FF manifest yourself to try it out.
The client side js code just picks up one in the list at random.
Also, looking at the js code, you can get the html to print a fixed joke by referencing the key in the json object on an "index" query param, ex: https://dadjokes736.onrender.com/?index=7
But I went in a different direction, it is a mix of RSS reader with summarization. https://rss.sabino.me/
It is open source, and hosted for free on github pages, so you can customize the feeds and reddit communities.
There is also a configuration ready to use the locall llama from github build system, so you dont have to rely on paying for AI services