I am often frustrated by ads/sponsored content on YouTube that I cannot buy. Youtuber present me nice product targeted for US audience. I am in Europe. No way I can use it or buy it. I would do it sometimes, but I cannot. Still I have to watch such ads.

I dont think there is a practical way to prevent this case.

That's the funny part, ads would be less annoying if they were hyper-targeted, which means there was more supply of ads and worse privacy. There's been a number of times I've found useful stuff from ads, but it's rare and almost never on Youtube.

Youtube is the one site worth paying for not to see ads and sponsorblock extension skips the live reads.

I am using this tool all the time and I did not know this! Thanks!

But files obsidian works with are just bunch of .md files that can be viewed or edited with anything, nano, notepad, visual studio code etc. So does it really matter it is or it is not open source?

How is your point relevant to the security risks of community plugins?

But also - no, they aren't, they use plugin-customized non-standard markdown format, so while the extension is the same, you can't view/edit them with anything just like you can't edit Word xml files with notepad (of course, it's not as extreme as Word xml, unless you're an extreme user of custom plugins)

It does if you care about running shady binary blobs on your system, and if you care about ethics.

It does if you care about your editor.

Yeah, but I dont see how your people can get away of ignoring laws of physics, as this is what the parent comment by "Zambyte" mentioned.

Wouldnt be better to have sandboxing built directly to cpython? Why there is no such thing already "include" in cpython? Or maybe to create some limited sandboxed venv?

I am often surprised how Claude Code make efficient and transparent! use of memory in form of "to do lists" in agent mode. Sometimes miss this in web/desktop app in long conversations.

I used to run in on raspberry pi, it worked flawlessly. Its like a hardware container ;-)

If you use LLM CLI tools like Claude Code you can let model just call shell commands directly instead of MCP. Or does MCP have some advantage even in the scenario?

no. it is mainly useful for accessing third party services via MCP. You no longer need to implement for each third parth API

I am also worried a bit about knowledge nowadays.

With LLM-based AI, should one also store individual chats in personal knowledge system? Yeah, I believe that some my chats are quite full of relevant info, that can be used in the future.

Also what is the right general approach here - should I ask the same question several times (every time I need information) or should I just look up previous answer in my history? To be fair I dont store google results, I just search it again, but with chat the path to right answer is often more complex than spitting few words in google search input box.

slight tangent: google's search results are abysmal, strong recommendation to try kagi instead.

With Obsidian, you have all your markdown files on your disk, so you can use vi or emacs to view and edit them while someone else put together replacement app ...