Redster

ChatGPT Advanced Voice Mode works fine, but Claude Voice is unusable, sadly. I'd rather talk to Claude! I would also like to try Gemini Live.

(I'm coming from the Svelte and Sveltekit ecosystem, which I also like.)

I left out that for almost every note, I try to link to or from another relevant note. If it's not relevant, don't force it. That just adds clutter.

But knowing future me won't remember {$relevant-note} unless I link to it helps me decide whether to link or not. If there's a note that know is obscure, but very important, I will sometimes link to it more than necessary.

1. It depends. I almost never reread for the sake of rereading. But there are notes that I keep coming back to over the past decade because I find the thoughts there worth revisiting or worth further development.

2. Like anything, if you put garbage in, you get garbage out. I don't find clutter generally to be a problem. I keep a couple of Junk Drawer or Scratch Pad notes for thoughts that might not deserve a note. If I create too many notes titled "Untitled 1", "Untitled 2", etc, I purge them or give them real titles. But clutter isn't a problem for me.

My biggest clutter problem has been including too many notes titled "{$DATE} - Meeting" or some other short, repetitive title ("Phone call with" or "Conversation with" are similar offenders). But I spend almost no maintenance on my notes unless I either can envision some specific usefulness or feel some specific pain point. And I rarely feel clutter in my thousands of notes.

3. I don't find that burden growing, no. I only alias or link if I see specific usefulness to future me. If I plan to write/think about a topic I might create a $TOPIC_NOTE and then I'll link all my relevant files to it, but again, those are often already linked to each other, so they're not hard to find and the effort remains small.

I don't have to maintain hardly anything. I spend almost zero time maintaining things and still use and can find notes going back to 2016 when I started taking notes in earnest and before I started using Obsidian.

4. I find full text search much more tedious than searching via filenames and links. Full text search is useful and necessary for very specific citations or numbers or names, if I'm not sure what note it is in. But for 95% of my usage, if search functionality disappeared from Obsidian, I would barely notice in my personal notes. I rely on search when I'm wading through other people's writing. I rarely need it for my own.

To ensure your future notes are useful to you, read what you have already written weeks or months (or more!) ago. Initially, I found my writing was too sparse. Too much context stayed in my head and my notes were almost useless. I reviewed them just before I completely forgot the context and updated them. The feedback loop of reading my own writing has helped me improve a lot at writing well for future me. I have to write to myself as if I will be an amnesiac in the future.

As for finding my writing when and where I need it, I try to give every file a unique and memorable title so I can refer to it in other notes and writings I access more frequently. For me, this means linking to notes and thoughts in Obsidian and creating aliases. If I know I wrote something down, but think the title is "Note B", but the title is "Note A", when I find Note A, I will add an alias with an additional title or keyword so that in the future, searching for Note B will surface Note A (and Note B, if it exists). Or, I will add a link to Note from the 1 or more of the most topically relevant note.

P.S. This maintenance doesn't take me much time. I only do it if I notice that a note was harder to find than it should have been and it usually only takes a few seconds to make sure it's easier next time.

(I used to use OneNote and Evernote, but bi-directional links with auto-suggest and a file quickswitcher are such gamechangers.

I was there when Obsidian and Roam Research felt like the first two options doing this seriously. The ecosystem has grown up around me and I love it. But you can do this with loads of other tools now.)

My filenames generally follow a "YYYY_MM_DD_HHmm Descriptive Title, perhaps a thesis statement - Source, if applicable" format.

As a case study, there is a Substack I follow with a guy who does original research. I've been following him before he started using AI images and once he started, I noticed that there was attention to detail in his images, in that they were highly relevant and unique to that article, sometimes including humor, and in general, you could tell he had actually thought about the image he posted. In his case, the AI art is mostly neutral for me. It doesn't affect my perception much one way or another.

On the done poorly end of the spectrum, when I go do a blog and every image is in the same style and is some variation of a robot or a person in front of a chalkboard (with the title or thesis on it) teaching other robots or people, I certainly think "Slop!" In that case, I more often feel that images are degrading content that might actually be decent and that no image would be better than this. I don't often come back, though.

I love NightHawkinLight's approach to practical amateur chemistry. He's different from, say, NileRed or NileBlue in that he's not working with exotic materials just for the sake of messing with exotic materials. Each video seems to be trying to grow in his understanding of chemistry for the purpose of solving a real world problem.

And props to him for running experiments and reading old papers!

Also, if this is your jam, the YouTube channel [How to Make Everything](https://www.youtube.com/channel/UCfIqCzQJXvYj9ssCoHq327g) is fun because the guy is actually trying to speedrun the human tech tree. He has some cheats that make it work, like once he "invents" something, he will allow himself/his team to buy cheaper/higher quality/modern versions of it, but I think that adds to the fun.

For actually sending, Signal disappearing messages or phone calls for some info.

I've really enjoyed Gemini 3 Pro. It's a good model sir.

Respectfully, A grateful dad who was homeschooled and who will homeschool.

P.S. Of course I will do some things differently than my parents, but it was an amazing gift and I had an extremely vibrant and stimulating time, including with peers (and adults!) outside of my parents' network who pushed me, challenged me, thought very differently than me, etc.

"So Gemini 3 Pro dropped today, which happens to be the day I proofread a historical timeline I'm assisting a PhD with. I do one pass and then realize I should try Gemini 3 Pro on it. I give the same exact prompt to 3 Pro as Claude 4.5 Sonnet. 3 pro finds 25 real errors, no hallucinations. Claude finds 7 errors, but only 2 of those are unique to Claude. (Claude was better at "wait, that reference doesn't match the content! It should be $corrected_citation!). But Gemini's visual understanding was top notch. It's biggest flaw was that it saw words that wrapped as having extra spaces. But it also correctly caught a typo where a wrapped word was misspelled, so something about it seemed to fixate on those line breaks, I think. A better test would have been 2.5 Pro vs. 3.0"

After continuing to use it, I genuinely think "It's a good model sir" and plan to add it to my rotation.

"Krita plugin Smart Segments lets you easily select objects using Meta’s Segment Anything Model (SAM v2). Just run the tool, and it automatically finds everything on the current layer. You can click or shift-click to choose one or more segments, and it converts them into a selection."

I hope to see this legislation in TN changed to allow cloudseeding.

I clicked on the "Technology" page (and all of the other pages, but there's scant info.

Gimp, Inkscape, Reaper, DaVinci Resolve, what else works on Linux to fill the void of Adobe?

(I hope to never give another cent to Adobe after how they treated me.)