it's the next paradigm in nostr that makes follow lists obsolete. instead, you will have thousands of data points which are weights given to other profiles based on positive or negative signal, and the notes you see will be ranked based on the weighted sum of interactions seen across all of your selected relays

the nostr algorithm will be perfect for this

eat the meat raw

rotate them

the onboarding flow in nostr is a crucial area of competition for setting up new users with relays that will accept their posts, holding their keys until they decide to burn the profile and self custody keys, and most importantly for providing a nostr algorithm and tuning model that is tailored to the user

I like their idea of forking algo's too. in my concept of the nostr algorithm, you will be able to fork algorithms as well as weighting models, and compose your own custom model from models of other profiles you trust, which could be human or AI

Currently, Nostr operates in a relatively simple manner, analogous to a basic neural network. The potential exists to make Nostr more powerful by:

- Creating more complex, layered algorithms (like a deep neural network) to filter and process events.

- Allowing users to define custom models or rules for how events are prioritized or filtered, similar to how deep learning adjusts weights across layers.

- Leveraging the interconnected nature of Nostr profiles to create a more dynamic, evolving system for event filtering and content management.

Nostr could evolve from a simple, direct event handling system into a more complex and adaptive one, where multiple profiles contribute to filtering and processing, just as multiple layers in a neural network refine the input data.

the nostr algorithm mirrors the approach of deep learning at a higher level of abstraction. it's as if we've all been tinkering around with a single layer neural network without exploiting the possibity of leveraging other nostr profiles to filter incoming events through a deep network of dynamically evolving algorithms and user specified weighting models

will check it out. I can get behind the no images ethos!

you're going to like the nostr algorithm

GM

I treat a boost the same as just another interaction for now

that might be a good idea! this new discovery mechanism does call for an overhaul of the typical client

based

killed it!

GM

This problem can be solved with an algorithm that evalutes each event by taking the weighted sum of all the interactions seen on that event, where a user model defines weights for a set of profiles. A positive weight means interactions from this profile increase the ranking of the event this profile interacted with. A negative weight does the opposite. All the events are sorted by rank, and the top results are shown for an evaluation over a given timespan. Previously shown events are exluded from subsequent evaluations.

GM 🤙🏼

GM 🤙🏼

GM ser