Discussion
Search code, repositories, users, issues, pull requests...
cyanydeez: no open code plugin? This seems like something that should just run in the background. It's well documented that it should just be a skill agents can use when they get into various fruitless states.
Grosvenor: Sparse distributed memory is what you’re looking for.https://en.wikipedia.org/wiki/Sparse_distributed_memory
the_arun: Aren't tools like claude already store context by project in file system? Also any reason use "capture" instead of "export" (an obvious opposite of import)?
gfody: yegge has a cool solution for this in gastown: the current agent is able to hold a seance with the previous one
kami23: Cool to see others on this thread.Here's a post I wrote about how we can start to potentially mimic mechanismshttps://n0tls.com/2026-03-14-musings.htmlWould love to compare notes, I'm also looking at linguistic phenomena through an LLM lenshttps://n0tls.com/2026-03-19-more-musings.htmlHoping to wrap up some of the kaggle eval work and move back to researching more neuropsych.
esafak: How does it select what to forget? Let's say I land a PR that introduces a sharp change, migrating from one thing to another. An exponential decay won't catch this. Biological learning makes sense when things we observe similar things repeatedly in order to learn patterns. I am skeptical that it applies to learning the commits of one code base.
pbhjpbhj: Memory links to location but that's largely because humans are localised. Isn't that also a weakness. We should be trying to exploit the benefits of non-locality [of ML models and training data] too.I feel like much of my life is virtual, non-localised. Writing missives to the four corners of the wind here and elsewhere; gaming online; research/chats with LLMs or on the web, email with people.My physical location is often not important - a continuing context from non-physical aspects of my existence matters more.That said, one of the things that's hard for me about digital life is the lack of waymarks - I used to be quite "geographical" in my thinking. Like "oh the part I found interesting was on the left page after the RGB diagram", I'd find that and also find my train of thought and extend it. Now, information can be in any myriad of freeform places across at least 3 devices and in emails, notebooks, bookmarks, chat histories, and of course my brain. When some ready syncretism of those things happens it feels like we'll make better advances. Personal agents can be a part of that.
swyx: hmm the repo doesnt mention this at all but this name and problem domain brings up HippoRAG https://arxiv.org/abs/2405.14831 <- any relation? seems odd to miss out this exactly similarly named paper with related techniques.