Oddbean

The problem with follower counts is skewed incentives Not only does it incentivize certain types of note formats (short, lacking nuance, provoking), but it also makes people regurgitate the same things over and over - leading to basically LinkedIn. People start using different clever ways of saying pretty much the same thing that doesn’t even need to be said in the first place. Of course, who decides what can be said? No one. So you get the enshitification of social media. I firmly believe that follower counts should be resisted as much as possible and social dynamics more closely mimic real life. In real life our reach is limited, and if one person says something that thing becomes forgotten fairly quickly. People do not gain 100 followers by saying one thing. They may build up some social capital and eventually a following if they stay the course with their message, but those are few and relatively harmless. But when you have everyone competing for a superficial score, this ruins the social web. People start competing, making the rest suffer. I mean, it would still happen without a follower count, but to a lesser degree without the underlying incentive. That’s all I have to say about that today. Maybe I’m saying the thing that doesn’t need to be said in the first place 🤣😯

I was searching for a video on tips for creating cooking videos and came across a video about how to get more Instagram followers. The game is to create content weekly based on whats trending.... Older content won't be on trend. Doesn't matter what the content even is... Just be consistent and on trend to get more followers. Create fiat. Get fiat followers. This seems like the down fall of society. #peakfiat

This def needs to be stated, and stated forcefully and stated often! Proxy indicators of trust like the follows count have SOME utility but you are correct to say we should resist relying on them as much as possible. Which means we need replace them with something better. And imho, that means: explicit contextual trust attestations, where you can say exactly what you mean and mean exactly what you say. Attestations of this nature are harder to game than proxy indicators. Which is why I’ve been advocating for them for so long. There’s a certain honesty about saying straight up “I trust Alice this much in this context.” No interpretation is necessary. My next step at brainstorm.ninja will be to enable attestations like “I endorse (npub) to write wikis in Category XYZ” and to see that reflected in the category-specific Influence Score. Which will allow CONTENT DISCOVERY of creators EVEN THE ONES WHO HAVE ZERO FOLLOWERS.

Communities can be "this context". They're centered around interests (context) by default. I'd filter recipes using the "Ray Peat" community I'm part of and family movies with "Based Homeschool Gang".

In what I'm drawing for communities UX, different types of content get surfaced first depending on how much they're used in that community. There's no reason why that cannot be used the other way around to automatically have a default filter anywhere: 1. User opens random recipe app 2. App checks for the user's communities that actually write/interact with recipes (in descending order) 3. App surfaces high signal recipes

I think this is the way... inferring amount/context! What I was referring to are explicit attestations, I think it's hard to get a good UX for a user to define how much and which categories. And I've seen alternatives mentioned like "oh no problem clients will deal with that" probably meaning that clients will sign events on behalf of users? If that is the idea I disagree

That said I'm in favor of a stronger trust signal. For example a kind 3 style list with "trusted people". But every single time we get back into the discussion of "trust for what" and "trust how much" - back to the UX problem. The latter is required for automatic trust assumptions - I'm yet to be convinced it's a good idea. It's like if on zap.store we automatically installed packages because "web of trust okay". No, I want to work algos that work hard to show you the most accurate information possible so *you* can make the actual decision.

“Trust for what” and “trust how much” is what @Lez ‘s proposed NIP-77 is all about. We’ve had some interesting discussion over the last day or two about how to represent context. Check out the PR discussion thread here: https://github.com/nostr-protocol/nips/pull/1208

While I agree with most of the thread, I wonder how will these attestations (context/score/confidence) be input into the system? I have a strong intuition people will not bother doing this. UX is everything. So yes, it may be a task delegated to an incentivized curator - which in turn requires trust in them. Which might as well be an LLM that ends up deriving trust based on proxies. IMO for this to actually work we need users to signal trust in a more progressive fashion. User prompts should be sparse, minimal and appropriate to the context of whatever they're doing in a client, e.g. when fat zapping Lyn Alden user could be prompted to sign an event assigning more weight to Lyn in the topics of the zapped note. A bit more, not a number. A perhaps extreme, terrible UX example would be: out of the blue, ask me in absolute terms how much I trust Lyn Alden, make me input what things I should trust her on, and the confidence I have on what I just wrote. Also, humans do not think about trust in percentages and are lazy when the ask is too big relative to an eventual payoff, so will answer things quick and inaccurately just to get done. It's a translation, a degradation in signal, so one could argue that these attestations are also proxies. Better than follows, probably, but also proxies.

I think you’re correct when you talk about doing it in a progressive fashion. Maybe you can help me select which of the following options would be the best user experience and make the most sense for my next update at Nostrapedia (at brainstorm.ninja): Option 1: enable thumbs up / down button for individual Nostrapedia articles. This will obviously be familiar to users. Under the hood, if you like an article written by Lynn, and she has placed it under the category of Economics (the author assigns the category as per NIP-54), and you are in my trust network, then I will see Lynn have a high trust score in the context of economics, even if she’s otherwise completely disconnected from me. Option 2: enable the “Follow” and “Mute” buttons on a user profile page, which of course users are also familiar with, but with an additional selector that allows you to follow or mute a profile *in a particular category*. You will select the category from the list of Nostrapedia categories, generated from all currently existing articles. I could even add an additional “superfollow” option, perhaps designated by a star emoji. Maybe you could even do a Superfollow with 1, 2, or 3 stars. In each case, I already know exactly how I’m going to interpret the attestations and process the data. Users won’t have to think in terms of numbers or look under the hood — unless they are curious, in which case they will have the *option* to verify all the calculations themselves. How many will be curious to know how the numbers are crunched? I don’t know. If they’re not curious at all, they can just keep clicking emojis like we all already do.

Here’s where Option 2 is headed. Suppose you have a normie friend you’ve been trying to recruit to nostr, who loves comic books and posts great summaries of his latest comic book finds. Normally he would make a nostr profile, look around, get no followers, see everyone talking bitcoin 24/7, get bored and leave. But you would create a superfollow for him in the category of Comic Books. The next day, your friend would have 10 new followers who all love comic books and saw your friend with one of the highest scores around in that category. And he’d stick around!

Yeah man absolutely we need to keep iterating and finding some usable middle ground. Anything with slight more signal than follows is a win. Option 1: would this be using kind 7 likes/dislikes? If so, trust would be derived from the interpretation of those events rather than being explicit, did I get that right? Regarding NIP-54 how likely is it that she adds the category? Option 2: love where it's headed, we need this. Communities fixes it too right? Today, how would we add a a mute event *in some context*? (reuse kind 10000 or does it have to be NIP-77?) In zap.store eventually I want to start asking, after a few app updates: "do you trust this dev" and make the user sign that attestation.

Probably yes about using kind 7 likes / dislikes. And a big fat yes to interpretation. I am increasingly thinking of NIP-77 as an idealization, with interpretation of proxy data into the idealized NIP-77 format as being a fundamentally important thing. Interpretation is done by the consumer, not the author of the primary data. It’s not far from what we do in real life! Maybe around half of NIP-54 articles are given a category, something like that. Eventually you’ll be able to invent new contexts and categories without the need for a wiki author to publish one, but gotta start somewhere. For a mute event in some context, I’m thinking I’d use NIP-77. But the beauty of interpretation is that if client A uses NIP-77 and client B uses an augmented kind 10000, the consumer can take data from both formats, interpret them into a common format, and pool them to create a larger dataset.

“This context” is a hard problem because we need an ontology of contexts. And ontology is a hard problem because we need SOMEONE to be in charge of the ontology, like a World Wide Web Consortium (w3c) committee of experts perhaps, but we simultaneously need NO ONE to be in charge of the ontology, bc the last thing we want is subservience to another goddamn committee of experts. What we need is a decentralized language, like the spoken word, but digital. **This is the hard problem.** And almost a decade of thinking about it is how I’ve come up with the tapestry protocol, i.e. the concept graph and the grapevine.

nostr:nevent1qvzqqqqqqypzqun2rcnpe3j8ge6ws2z789gm8wcnn056wu734n6fmjrgmwrp58q3qy88wumn8ghj7mn0wvhxcmmv9uq3vamnwvaz7tmjv4kxz7fwdehhxarj9ehx2ap0qqstkq8jw602y9s3m92vk0yujl87k3tspr38y8l6z80qmf0tk3p2y2sf58gh0

agreed, also, in the long run i think people will choose transparent algorithms to connect them to people/ideas they like rather than pure chronological follow feeds, so at best following is short cut to train your own algo on what you want