vid

I like LocalLLaMA because it's not just for algorithm/math experts, it's for people who want to use LLMs locally (as the name says), so there is often practical discussion on how to use "off the shelf" models and kits. This also implies a patient crowd who will sometimes go into great detail about specific details, since there are many non-specialists.

I think more than anything they need advocacy and good quality product contributions (support, documentation, code). From what I know, a lot of the development happens via a few consulting firms that support their larger clients.

Ok, but how could we really trust these companies? I'm not saying to "give up," but aside from "pinkie swear we'll never make a mistake or exploit a loophole again," how can companies be trusted with the most intimate details of people's lives (and the lives of people who never signed up but get dragged in incidentally), such that it becomes trivial to deceive or even obviate or impersonate them? And how to get away from this being normal, so every company with an IT department or cloud subscription has their own spy agency? Especially since even deletion policies may let them hang onto data for years, in which time divisions can change hands or laws change. And also, they've become embedded in state security policies as a sort of branch of the secret police.

I've been looking into NextCloud to recommend to a government agency. The world desperately need competition or at least something compatible with the m365 stack, because it's eating the world and taking a lot of choice away and killing a lot of innovation outside the Microsoft funnel, since Microsoft is not interested in a lot of tech (for example, network schemas, useful for "tell us once" type applications, since they'd rather you just use their tech for everything, and the messier it is behind the scenes, the better for them).

Anyway, I have mixed feelings. I admire the community and the support it has by many governments, its staunch Open Source basis so it's useful for an individual or a large organization. But it is building on a lot of crufty PHP, their collection of apps is very uneven and it's hard to know what works well without a lot of research, and it's going in a few directions to upgrade. AppApi in particular is on one hand very innovative, on the other going in some odd directions. I know it is successfully used by very large organizations, but without spending a lot of time with it, it's hard to get a sense of the commitment and considerations required.

"Power users" shouldn't just mean more little boxes on the screen. A 2024 client should offer in-depth background analysis, embracing the open nature of the network. Provide helpful context on people and topics within threads – be transparent about the information that's already out there. Visualizations, metrics and network detection features would make the client truly powerful, offering users transparency and letting them better understand what's happening for more control over their data. Flip the script on the centralization of insight and let people organize at a higher level.

Wow, some pretty strong opinions here. Did somebody say Nazis?

Tech innovations are always a mixed bag, but there are some great ideas in this, like selective disclosure, where you don't have to reveal all your personal info (address on driver's license) when proving your age. Is uploading a picture of your physical credentials to pretty random web sites supposed to be a better solution?

I also think people deserve to have well described data and interactions, and their "wallet" should become a coherent place of organizing long term information. These standards create that possibility. Letting companies dominate and invent things to their needs as they go along is not a better solution.

It takes some strength from governments to do this properly, which some will point to as backward (witness the reaction to iPhone AI features). But companies have not found the end point of innovation, it needs to be kept open and directed by broad people needs, not the current shiny.

I am not sure there are many significant science breakthroughs that aren't basically built on a 'full communist' (publicly funded) model.

There is a very significant tension between making all works that are produced on the back of giants all the way down free (practically speaking, everything, unless there are significant works developed by feral people), vs keeping individuals fed and happy, vs giving corporations so much power they only serve themselves.

I think the benefits of saying any significant development must at least publish its information metadata including descriptions has too many benefits from any perspective to not be supported, and I'm not sure it'd be that expensive if its incorporated in existing systems. It would create its own network effects.

What was really hurting me was all the literal strings "even" and "odd." I would have made them into constants. I guess then the next step would be to use constants for all the numbers.

I am probably going to lose some precious poinnnttts for this, but I sure wish they could instead target browsers instead, whether by using the canvas, WebGL, three.js, A-Frame, or something else. This would have the immediate benefit of making the software work on the Web or as a local webapp any device that supports a browser, and they would have a much larger pool to hire from. They could also come up with with great libraries.

It is not quite in the same league, but we already have SVGEdit[1] as an example. as well as diagrams.net, but taking a 2023 approach would be much more performant and open ended.

1. https://svgedit.netlify.app/editor/index.html

Someone should tell the author about the great lshw, it lists full details of the system board (and everything else) without having to go to a GUI display.

  azalp
    description: Desktop Computer
    product: B660M AORUS PRO AX DDR4 (Default string)
    vendor: Gigabyte Technology Co., Ltd.
    version: -CF
    serial: Default string
    width: 64 bits
    capabilities: smbios-3.4.0 dmi-3.4.0 smp vsyscall32
    configuration: boot=normal chassis=desktop family=B660 MB sku=Default string uuid=dotdotdot

People posting those CPU and SSD screens shots rather than, say, some linked data that contributes to make a graph to forums has such a deleterious effect on people's ability to reason. I suppose eventually we'll end up screen scraping the screen shots.

Sure, embrace the future. But it looks like something out of Mad Max, attenuated with the macho looking guys piling out of it. It has a very "me against the dangerous world" tone, something Musk has expressed before (in denigrating public transportation). It's a very aggressive introduction, highlighting the worst aspects of private vehicles, and a very poor direction to set.

When I was growing up, Tolkien's works were some of the few books sold in the church's bookstore. As I've grown older, I've grown more critical of the simple narratives in most literature, that make frequently abhorrent attitudes and acts possible. In the case of Tolkien, he literally dehumanizes the enemies. One example that stands out is the "light hearted" sequence where Legolas and Gimli hold a contest of who can kill the most orcs. These days, recognizing this type of storytelling, with all its justifications, is very often part of the problem, I try to find works that help develop more nuanced perspectives.

Edited to remove bad reference to Nanking Massacre. The point is that glorious battle stories with simple morals are prevalent, and are a big part of Tolkein's stature.