Woah I have the same laptop stand arm at work

What services do you prefer? Ive no dog in the fight because I just use the CLI but im curious what’s out there

There are some android TV devices that you can run LineageOS on which is what I do. But they are few and far between, and require specific firmware versions which makes it pretty impractical. I have used a Raspberry Pi with LibreELEC before as a frontend for Jellyfin and that worked decently well.

My dream is for Plasma Bigscreen to be supported well enough that I can just use a standard Linux machine for my TV, and there are tons of options in that ecosystem for casting, airplay, etc. But last time I tested it the issues were too numerous to ignore, specifically around application stability and UI navigation

I think it’s a pun about “stables”

I have some GL. iNet routers and I like them a lot. Comes preinstalled with their flavor of OpenWRT but you can flash it to the standard image too. And they’re affordable (at least when I bought in last year they were)

Neurons are not an i/o black box. They have internal processes, transfer synapses in many directions, have receptors for myriad neurotransmitters, and the patterns of synapses communicate information above and beyond what one neuron can do. The system is greater than the sum of its parts. ML training, backpropagation, gradient descent etc is nothing at all like how neurons communicate. it’s nothing like thinking when the statistics of an LLM make a prediction through inference, and LLMs cannot “learn” in any meaningful way

Are you serious? We aren’t just prediction machines, we have thoughts, internal worlds, consciousness. We learn from new information and can envision what different possibilities would mean for the world around us. We reflect and introspect. We learn from our mistakes. We can creatively apply our knowledge, experiences, and ideas to make new and original things. An LLM has NONE of these core aspects of human thought and intelligence

All an LLM can do is remix it’s training data in different ways, it’s nothing more than a statistical prediction machine. It cannot do anything truly novel or original, so by its very architecture this notion of self-improvement is impossible

Slop

That still helps the rest of the world because it reduces demand on the existing oligopoly of memory manufacturers

But also, that number is a lie. They haven’t published any releases from that rewrite, it’s still considered experimental, and they’ve spent weeks continuing to iterate upon it and certainly driving that number way higher. But they declared it done via blogpost to get that price-point and timeline in the headlines. Yet it still isn’t shippable, and they’re still burning cash and time on it

Tailscale is pretty straightforward because they have a config for it baked in, pretty sure it’s just services.tailscale.enable = true. But if you use it with the operator=$USER setting, note that you still need to run tailscale with sudo for your initial login before that works properly

This whole community is running in <8 GB of RAM? 🤯 love the work y’all do, keep it up!

I’ve heard good thing about Jitsi, but I haven’t used it outside of matrix calls

FOSS and copyleft: I want software built for the good of the public, with good intentions. Anything with corporate licensing us off the table, and ideally so are non-copyleft open source programs as corporations will hijack them for their benefit and privatization.

UNIX Like: It is my preferred workflow for development and everyday use. The windows paradigm is messy, convoluted, and inconsistent by comparison.

Fast, free, customizable: all nice to haves connected to the big two above

Lenovo is the largest laptop manufacturer in the world by market share, and they’ve offered Linux preinstalled on many laptops and desktops for at least a decade

I use this layout somebody on their GitHub made that is similar to the IOS keyboard. I tweak it a little but it’s a good starting point: https://github.com/HeliBorg/HeliBoard/discussions/2180

I just use the native android private space profile, good enough for my use case

I’d be interested if they had WiFi calling, RCS, and other basics implemented. Hard to say how true the service is to its claims, but the feature set sounds compelling

I’d use the term Machine Learning to avoid the baggage of what AI implies/promises. But ML has tons of real world applications that other tech can’t solve. I view it as the best tool for fuzzy or approximate data. It’s used in statistical modeling, which can be applied to the automated generation of potential new drugs or proteins. ML can be used in image processing, with image recognition being a huge use case, as well as real-time upscaling like DLSS. And it can be great for Speech-to-Text, natural language processing is a difficult problem to solve but ML approaches do it fairly consistently. These are applications that can generally tolerate the sizable margin of error that a statistical model inherently possesses, either by not being mission critical or by being designed to get reviewed by human experts; I don’t think ML should ever be applied to areas which need guaranteed precision or accuracy. Also note that of all of these examples, only one uses an LLM, and none vie for the title of “AGI” or claim consciousness or intelligence lol. So that’s why I avoid the term AI. ML can imply some of the same stuff but it hasn’t been buzzwordified the same way