Nostr is better than Xitter, but it still favors narcissism over merit. V4V doesn't reward the best builders, it rewards the best salesmen. There is no centralized algorithm pushing a narrative from a continuum of a billion options, but neither is there a defense against constructed consent. The protocol can't stop the manufacturing of "hero figures" because we are biologically wired to seek leaders and follow them. I liked Reddit because it was about ideas, not people. You followed a topic and the design almost hid the author. This gave a sense of "Message, not messenger". Nostr is the opposite. It is entirely messenger-centric. This is evidenced by the flood of "GM" and "GN" posts, which would be completely void of value if it wasn't about who was posting them. How could we nudge Nostr away from this? Some clients allow you to hide authors, but that is just self-constraint. I think we need a protocol layer that constrains participants to focus on ideas but without getting drowned by AI slop. How would a "late reveal" of authorship work? Authors could prove group membership - being one of my follows and only after a week or so reveal their identity. Fundamentally I wonder if people want to be heard for who they are, or for what they say? Do I want to be famous, or do I want my ideas to be recognized? I think the latter comes first. We want to be recognized as authors of great ideas.

Who's whitenoising? It's a bit cumbersome to bootstrap. Either I put my nsec into the app or I won't find my follows? It would be nice if I could provide my npub leo@nostr.info to find relevant others.

When I thought about the AI apocalypse ten years ago, I thought we'd be fine as long as the AI doesn't get internet access. Now it gets millions of root servers, credit cards, bitcoins, ... to play with. Are we dangerous?

Why do I see this on Amethyst and not on Yakihonne? I even re-broadcast it. View quoted note →

Vibe coding is controversial. @jb55 claims to code features as fast as he can think of them with the same quality as hand-coded and others claim to be slower vibe coding than hand coding stuff. The truth is probably in between and my experience is that it's easier to get going on a challenging task but ultimately it's a lot of cleaning up after the AI, so maybe I'm faster cause I get started at all but then, cleaning up after the AI costs a lot of effort and often results in me abandoning tasks. That said, the very near future will probably be "coding as fast as I can come up with features". Maybe it's only missing a context window big enough to grasp the entirety of the project to hook the feature in where it belongs and use the tools that are already implemented ... or at least use the programming languages already used in the project. But I suspect, a bit more than token count will be needed but it will come within a year or two.