_matthew_
@hypfer I'm curious, can you understand / identify with at least some of what I'm sharing below?:
A lot of people (likely near 100% of people) have their own internal beliefs about the world and they are largely commenting primarily to validate that belief (I am literally doing it with this comment). Like you often can see both what the person objectively wrote but then also get a feeling for some internal belief that motivated this specific comment.
And naturally, people are emotionally attached to their internal beliefs. And so sometimes comments are less a sharing of ideas and more a factually coded emotional rationalization of their internal beliefs. And when you read a comment like this that goes against one of your own beliefs, it may feel like an attack.
To me the distinction though is the motivation of the person on the other side. My interpretation is that true trolls have a motivation to incite emotions whereas most other people simply are trying to validate their own beliefs, despite potentially doing it in a way that is not emotionally palatable to all perspectives.
Separately, for you (hypfer), do you agree that there could be some amount of "hostile attribution bias" here? I saw this term while searching about this topic and to me it would explain both your perspective of feeling personally attacked and others perspective of feeling like that's actually not at all what's happening.
Regarding swapping, are you sure it's not just from ram backed tmpfs partition? For me that was it and most of the swap was from /tmp. So I created an agressive job that runs frequently to delete all tmp files not used in the last ~2 hours without any command holding handles to it.
The other thing I found was that there were a bunch of leftover processes from prior coding agents forking commands and just leaving them. So I had to build a system into my agent spawner to automatically launch each agent session in a cgroup whose processes all get killed on task completion.
Like, the "modern" electric cars of today are heavily inspired by the common portrayal of "futuristic" tech from the years prior (think the electric "engine" sound used / mandated now). Conversational AIs that basically have recreated Her (2013). Even the personas that chatbots take on I would make the case are drawn from the actual characters in countless stories in its pretraining text.
And so, I would argue, that the more stories we have of optimistic futures where AI integrates into society, the more stories existing AIs trained today have to draw from to actually tune their behavior to. And similarly, the more positive futures there are to be inspired by, the more we may see investments to build products leading us to such a future.
In fact, if LLMs begin automating large sectors of innovation, the actual directions of products and research direction they take may literally be influenced directly by these stories.
To me that's the crazy part.
As an aside, technically, yes, it may have been better to converge on some more clearly "fake" unit rather than calling it a pixel but for me that's a nit pick.
However, if you just think of these kinds of LLM driven software factories as having a limit in terms of complexity (ie. some asymtotic / logarithmic increase in cost beyond aome complexity threshold), it still is quite a useful system to have (despite being admittably more niche) because it lets you quickly explore various product ideas, user experiences, and featuresets.
My personal take is that the best approach is to first prototype products as vibe coded slop using a software factory like this to understand the user experience, featureset, and product concepts. Then, once you have a clearer understanding you can properly design out the system and more carefully rebuild it either from scratch with a more costly model and clear architectural details or strip out the core systems and redesign them in large refactors.
https://github.com/matthewscholefield/blobse
Which powers a few small apps like https://matthewscholefield.github.io/votosphere/
Have you considered adding an llms.txt for linkedrecords? I'd love to be able to just tell my agent to build something and point it at your llms.txt and have it just build it out on its own reliably without having to clone the repo, go through the docs, etc.
Of course though they are not necessarily a viable solution for companies with security requirements etc. given it is just a single person project, but they still serve as a proof it can be done.
It's probably pretty niche complaints but I didn't like how overly precise due dates were (every due date has precision down to the second), how recurring tasks worked, especially with cross device synchronization, and a few other small things.