Home
Artists
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register
Partychan SimpleX Chat ThePornDude
Home Artists Posts Import Register
Join the new SimpleX Chat Group!

Will Linux be left behind because of AI & Channel updates - Patroncast S03E49 (Patreon)

Published:
2023-12-11 10:21:10
Imported:
2024-03
Tags:

Downloads

Content

Hey everyone!


I hope you're all doing well, because here is your fresh weekly rant!

In this one, I talk about:

- 00:47 The channel, the current financial situation and the worries I have about sponsors, and what I'll be working on

- 11:20 AI integration in proprietary OSes and apps, and how this might impact Linux and open source software in the long run: will we have to add AI powered features to our desktops and apps to stay relevant, or will we refuse to, and be looked at as the "primitive" way of using a computer?


I hope you'll enjoy listening to this one, and I hope you'll have a great week!

Best,

Nick

Comments

Emil Johansen

Separately: Will it work well on Linux when someone implements it? Your answer is in your previous patroncast. Core to LLMs and deep learning / neural networks implemented with high levels of utility is a strong parameter space / input enumeration. That obviously does not make a solution on its own, but without it you get much lower utility and more toy factor - like companies slapping an OpenAI API on a site with a bit of pre-prompting and calling it a day. So essentially you need good points of measurement. For output you have the LLM capability of contiguous streaming, so you're generally always good there for any concern about output resolution. How does that relate to the last patroncast? Well, you talked about Wayland - and more importantly portals. Combined with systemd and similar trends, these provide an excellent common interface for all applications & processes on a Linux desktop. An interface which could be tapped for measurements fed to a locally running personalization knowledge base. Essentially a type of system log, but written for database storage rather than spelled out text messages. Essentially this is what Siri on iOS has been doing for years. It made a lot of sense to implement there as, like portals and systemd on Linux, iOS features a strict access framework through which apps interact and thus provide points of measurement. For what I have seen, Siri just hasn't _used_ that database for all that many interesting things, but the basis for something with high utility is there. This also explains why we have not seen something like that on Windows before. Unlike iOS and the modern Linux desktop, Windows does not really have a good framework for measurements used across all applications on their platform. They are going to need that for this new project of theirs, but on this front they are starting from the back of the group. Even if they launch something nice, it is going to take a _long_ time for broad utility across a meaningful number of common platform applications. Not an issue which the implementers on the modern Linux desktop will face anywhere nearly as badly. But when the time comes and people have gotten used to more contextual smarts and personalization in their OS-diet from Windows world, they will find at minimum as strong options in penguin land.

Emil Johansen

Oh, re: "we are doomed if it requires cloud": Certainly, but it will not do so. The primary reason current players offer their AI solution through cloud API is nothing more than the classic: They would rather rent you a SASS solution than sell you a local one. Now what is more interesting on this tangent is we would most likely _want_ to have a degree of optional cloud supported. Not for running the system, but for carrying our personalization across systems and platforms. So the personalization from your work desktop is available on your personal laptop as well - securely and privately transferred via your nextcloud or whatever and properly contextualized & categorized/subdivided. So behaviour on the local machine takes precedent, but the fallback is more informed - and perhaps new habits / trends are taken into account or whatever. Or how about a web app in your browser requesting a portal of a certain subset of your personalization database - for personalizing its behaviour just like a desktop app would?

thelinuxexperiment
>>124421550

Hmm that's really interesting, I hadn't looked at it this way. Some sort of abstraction layer that would "secure" the data to only let its results be accessed through a portal would make a lot of sense: the data is there, but it stays private and aggregated in a way that's still usable by various apps.

thelinuxexperiment
>>124421908

There is that issue of "how do you keep this working across devices or when you migrate to a new one". Since these big actors are all banking on their ecosystems, it makes sense for them to be cloud based as much as possible, to sell this as a feature that only works on their respective platforms. i guess we would be able to replicate that with the current tools we use to replace these proprietary ecosystems. it wouldn't be a lot of extra work compared to the steps we already have to take to get the same level of "personal cloud" integration in Linux distros.