Initially about 15 years ago because the laptop I bought did not have kernel support for the mobile gpu so I could not run a graphical desktop for a while, so I added kernel support for framebuffer on my GPU myself which was only a 7 line change (upstreamed to kernel) and then realized the framebuffer plus tmux and I was kind of set apart from youtube, so I wrote a one liner to pull raw video content from youtube and stream to the framebuffer with mplayer.

Overall I found there was almost nothing to do my job that actually required a GUI so did not bother fixing it.

The only software I tend to use today is a terminal and a browser, and graphical browsers are a hellscape of cookie approval prompts and ads and tracking. Thankfully today a local LLM can go to the web and fetch whatever content for me, so I am once again phasing out my use of web browsers.

▲

negura 4 days ago | parent [-]

> The only software I tend to use today is a terminal and a browser

i'm exactly in the same situation. for this reason i run on dwm and no DE. but i'd like to phase out some of my web browsing, for the same reasons you mentioned. could you please elaborate on your local LLM setup and how you use it to substitute web browsing?

▲

lrvick 4 days ago | parent [-]

"Can you setup a daily job to monitor and download youtube videos from any channels listed in subscriptions.txt"

"Can you setup a daily job to search for any articles on supply chain attacks, format them as normalized markdown files and store them under Research/supply-chain ?"

"Can you go find me a good price on some used DDR5 ram from a reputable seller?"

Etc etc.

Use agents to bring the internet to you.

Can do basically anything with a Strix Halo board at home.

	▲	negura 3 days ago \| parent [-]
		interesting. i can definitely think of similar use cases. but can all of this be accomplished with local LLMs? do you run it on expensive hardware? are there specific models, agent frameworks and other tooling you recommend? btw as another strategy to avoid web browsing i've been relying a lot on content being delivered as email newsletters. seems like it would also be easy to process that using LLMs