Sims

joined 2 years ago
[–] Sims@lemmy.ml 3 points 9 hours ago* (last edited 9 hours ago)

You have to think in 'information dimensionality'. A yes/no toggle is 0D, a list is 1D, a list of lists (std hierarchy) are 2D, a list of list of list are 3D etc. All information storage types are one of these dimensions. Think of a graph-base file system with nodes and edges between everything. Now, imagine a filesystem where you flick a switch and the whole structure shows another pov ? Maybe you want the whole thing to be shown as file-type hiearchy, or only parts of it. Maybe you need to show movement in the structure, so everything are in a temporal/spatial hierarchy, maybe you are only interested in dependencies ? Relations ? Other 'weird' metrics ? ..and so on. The main problem is to manage, find and show the needed information in a higher-dimentional fs.

Technically a normal file is also a list, or another ordered structure, but in this sense, they are just a node with further dimensionality.

There's a TON of information layers locked away in our normal filesystem hierarchy, so OP are perfectly right, and people here have no imagination or even a world model of information structures..

[–] Sims@lemmy.ml 1 points 9 hours ago

Not sure why people defend an archaic organization form here - reflex ?

You are perfectly right that files and folders are simplistic, and should naturally adapt to the pov that are more information rich/valuable. Hoomans tend to collapse a high-dimensional structures to 0D to 3D, so we can manage the information. In that sense, a std hierarchy is only ONE pov over a ton of pov over the same content. A standard hierarchy is only a low 2D dimension structure that are our first attempt at organizing information. It's not wrong - just imprecise af.

Anyway, hardlinks are a small step up, can build wild static structures (like a oneshot filesystem in Guix), but is cumbersome to control in multi-dimensional information structures. Likely not what you want, but look into fuse file systems if you want to move on to a dynamic file system hierarchy. An interesting one is a tag file system. It turns a standard limited hierarchy into a much more dynamic file-structure where a file can - and does - belong to a bunch of tags - file type, size, group, comments, whatnot. There are many many fuse fs that can convert anything into a better structured file system. Tagging is a step up from a dumb 2D hierarchy, but maybe a graph file system is the ultimate freeform dynamic filesystem that can present all the pov's we could possible need ?

Go for it.

[–] Sims@lemmy.ml 7 points 1 week ago (1 children)

Nonsense propaganda from the US tech-lords/Plutocracy. Who gave big tech the power to investigate and convict anyone - without any oversight ? Ridiculous warmongering..

[–] Sims@lemmy.ml 4 points 1 week ago

Alternative idea for the adventurous lot - or extreme self-hosters: Buy an old Xeon datacenter server with 1.5tb ddr3 ram (separately). Ddr3 still costs around a dollar pr gig (a month ago), so this rig with two xeon and 24 * 64gb ddr3 ram will cost ~1900$ + transport: ~2k.

(OR an older pc motherboard with 4*64 channels or similar)

DDR3 is a bit slower, use more power, and the server is big/noisy, but the difference in price is unbeatable, and it's a hell of a KVcache (for ai) if you put a gpu in it.

Note though, that power expenses are going up in the West.

[–] Sims@lemmy.ml 3 points 1 week ago* (last edited 1 week ago) (1 children)

Not helping here, but I heard a guy with Guix did that. Guix just builds a profile with the extra desktop parts, run it in a local container if you want and add that profile to the local software stack. Not a vm but maybe you don't need it ? Both the system, home and the desktop profile are declarative, so very mobile. I think he had his DE user profile remote also, so extremely minimal/air-gapped and stable solution with almost zero local data: system, home, desktop-profile, remote user profile (ldap etc).

In declarative operating systems, you describe what you want, and the system builds it for you. Your whole system configuration is a few files of std code (learning experience ;). Personally, I'm done with the usual monolithic distros. They are too error prone for my taste, and not really moving with the dev flow of operating systems imho.

Anyway, just a loose rumor/idea, I have no links and don't actually know how to do it, sorry.

[–] Sims@lemmy.ml 32 points 1 week ago (4 children)

Yeah, I lost my girlfriend, so .. no judging please !

[–] Sims@lemmy.ml 5 points 1 week ago (1 children)

How unfortunate to speak 'AI', when people are currently scrambling to build 'AI detectors' that removes AI speak from their feeds. Good luck buddy 😵

[–] Sims@lemmy.ml 2 points 1 week ago (1 children)

Great that you chose weights for the final evaluation. Black, and whitelists are just too ..0D/binary for any advanced use, and are unfortunately used everywhere.

[–] Sims@lemmy.ml 8 points 1 week ago

Yes, every neo-liberal ceo they have had, are dragging mozilla towards a full commercial strategy, while loosing market shares af! I don't think US have many non-neo-liberal leader types..Anyway, they and their anti-community attitudes have the full responsibility for the multi-year dive.

[–] Sims@lemmy.ml 1 points 1 week ago

No, you can access something like 'localhost:8080', so if you have your ollama/webui or other agent/llm listening there, FF will show that interface in the side window. Summary etc still works. You just chose 'localhost' as your provider.

Better search, but it looks like this: browser.ml.chat.provider http://192.168.1.83:8080/

[–] Sims@lemmy.ml 7 points 1 week ago (2 children)

Everyone’s sick of AI

But not everyone are ! But a group of people have gone to the 'hate the tool' corner, are very aggressive and feel that 'everybody else hates it'. But 'we'/'they' don't - some do.

Any software/tool that leaves it up to the user if they want to use such a feature or not, are okay. Open source and Foss doesn't exclude the use of AI, and 'catering' to a special group of angry people that just want to remove the option entirely for everyone else, is not very 'community' aligned in my view.

There will be plenty of clones that pull out this or that part of the official source tree, so nobody will lack a normal browser. I currently use 'Zen' and I had to 'force' AI on. Other will just remove it entirely.

[–] Sims@lemmy.ml 1 points 1 week ago (1 children)

More primitive anti-china propaganda. Who believes this thick nonsense ? The western Plutocracy are really firing it up ! Imagine that the fascist techlords are now suddenly 'trustworthy' good guys - and joining up with 'national security' in a little ww3-mongering propaganda campaign. I'm sure there's no cheating and the evidence would hold up - if anyone ever got to check it out oc.. tsk tsk..

 

I thought of this rare little sub when watching this. It's difficult to evaluate how much Yank (and western) culture have contaminated our expectation and knowledge of other cultures. This video explores some of that..

And, ..going slightly OT on my own post, when trying to search for African music, the result contains mostly African 'beyonce-like' music. I have to add 'root' or similar to find something that actually sounds 'original', local and non-yank. A bit sad.

 

I am planning my first ai-lab setup, and was wondering how many tokens different AI-workflows/agent network eat up on an average day. For instance talking to an AI all day, have devlin running 24/7 or whatever local agent workflow is running.

Oc model inference speed and type of workflow influences most of these networks, so perhaps it's easier to define number of token pr project/result ?

So I were curious about what typical AI-workflow lemmies here run, and how many tokens that roughly implies on average, or on a project level scale ? Atmo I don't even dare to guess.

Thanks..

 

I could not tell that they were generated..

 

Hi all.

For a long long time I've been very happy with Signal, but have lately become rather annoyed that:

  1. it too often bugs me about about an update and forces me to do an update before I can write to my single/only recipient, and
  2. it too often bugs me about my pin code, even tho I never asked for such an annoying level of security.

These security measures are completely overkill for my/normal use, unnecessary, annoying and very aggressive. I'm an adult, and unless there's a super dangerous zeroday attack/vulnerability, I don't need constant forced updates, I don't want to retype a pin code for any reason or interval, and I certainly don't need to be told how I should run my system, when to upgrade or have software on my system that 'randomly' gets locked down for whatever reason.

Does anyone know how I can turn it off (Linux, Android) ? Is there another client fork that don't force me to follow their idea of what security level is necessary ?

Thanks, and apols for negativity..

view more: next ›