I like the #smallweb the way it is. Quiet, a little rough around the edges, not trying to sell me anything every five seconds. It reminds me of how the internet used to feel before everything got optimized and polished into the same shape.
That said, there are a few things it could use without losing that spirit.
First, better ways to read. Not fancier, just smoother. If I’m following a handful of gemlogs, a couple Mastodon folks, maybe a Lemmy thread here and there, I shouldn’t have to juggle three different apps and a dozen tabs. Give me something simple that pulls it all together and lets me just sit and read. No ads, no tracking, no friction.
Second, a little more durability. Too many good sites just disappear. I get it, people move on, but it would be nice if there were easier ways to mirror or archive things so the good writing doesn’t vanish overnight. The small web has a memory problem.
Third, discovery that doesn’t feel like an algorithm breathing down your neck. I don’t want “recommended for you.” I want “here’s what someone else thought was interesting.” Old-school blogrolls, human-curated lists, maybe even random links that actually surprise you. Let people point to things they care about, not what performs well.
And maybe this is just me, but a bit more cross-connection wouldn’t hurt. Gemini, Gopher, the web, the #fediverse all feel like neighboring towns that don’t always have good roads between them. You can get there, but it takes effort. Smoother bridges would go a long way.
None of this needs to be big or complicated. In fact, it shouldn’t be. The whole point is to keep things human-scale. But a few small improvements could make it easier to stick around, read more, and maybe even contribute something back without feeling like you need to build a whole platform just to say your piece.
That’s really all I want out of it. A place that’s simple, a little more connected, and worth coming back to at the end of the day.
@queerlilhayseed@piefed.blahaj.zone Thanks for the positive response! You basically named the two hurdles that ate most of my development time on my projects system.
On the counting/nondeterminism thing: the trick was giving up on the LLM as a "knower" entirely and treating it as nothing more than a reader. So if I want to know how many times grass shows up in some sprawling Tolkien-esque passage, I'm not asking the model "how many times does grass appear" and hoping for the best. I use it to pull out and tag the relevant sentences into a structured database, and then a dumb, boring, deterministic Python script does the actual counting. Temperature's pinned at 0, and I force it to hand back verbatim quotes with citations instead of letting it synthesize a summary. That alone killed most of the hallucination problem.
The "systemic wrongness" side (bias, basically the model just inventing stuff) came down to hybrid search, meaning keyword/BM25 alongside vector embeddings. Pure semantic search will absolutely whiff on a weirdly named character or a specific term just because it doesn't "feel" semantically close to the query, so having the keyword layer as a backstop matters a lot. And I'm strict about grounding: no direct source quote from retrieval means the system says "I don't know" instead of guessing. It's not allowed to fill gaps with vibes.
Your instinct about cataloging prompts was right on too. I keep a golden dataset of queries I run every time I touch the system, and rather than some vague overall correctness score, I bucket failures by type: was it a retrieval failure or a synthesis failure. That distinction alone tells you whether the fix is in your chunking strategy or your prompting, instead of just flailing at both.
Honestly the tools have come a long way since you last looked at this stuff, and I think it's mostly a philosophy shift. People stopped treating LLMs like databases and started treating them like interfaces sitting in front of one.