Oddbean new post about login | logout So can you think of seperate problems that need to be solved in order for this to be (partially) solved?
Many. Here’s one: The Grapevine calculates an influence score that is context based — eg, Alice’s Grapevine tells her that Bob is awesome at rating movies in the comedies genre, so he gets a high score in that context, which means if she’s looking for a comedy to watch tonight, his ratings of comedies are weighted heavily. Problem: how to calculate the Influence score in a way that doesn’t give undue influence to Bob just because he has a zillion followers? The Grapevine’s solution: https://habla.news/a/naddr1qqxnzdes8q6rwv3hxs6rjvpeqgs98k45ww24g26dl8yatvefx3qrkaglp2yzu6dm3hv2vcxl822lqtgrqsqqqa28kn8wur
I believe you are talking about content rating, but this problem is an even more to the ground problem. Kinda a fundamental one. I think the problem is in the infrastructure itself. The most basic low-level stuff of how peers can know where to connect to beforehand to get certain data. in this case: how does Alice even find bob and his content every day, in a decentralized unstructured (or minimal) manner?
The way the Grapevine works, trust is contextual, and contexts are categorized, with trust inherited from general to specific categories. Example: User A trusts User B in all contexts; B trusts C on medical topics; C trusts D on pulmonary topics; D trusts E on treatment protocols for pulmonary tuberculosis. Each of these links is an attestation stored in nostr, and this is how User A finds User E and benefits from E’s expertise.
That is a really nice concept. Is it also implementable without tying it to individuals?
You mean could I make an attestation regarding a pseudonymous user account? Or an AI? Or a group? Yes to all!
Is it possible to implement this with individual data, (interpret data based on ...) instead of with an unique entity?
You will be able to design rating templates where the things being rated is a piece of data, and the fields are whatever serves your purpose. Maybe the rating template is a flag that the data is true or false, or is (or is not) NSFW, or is / is not supported by randomized clinical trial data, etc. And you could attach a confidence rating so you can indicate how sure you are of your rating. And a comments field, if you like. And if the community likes your template and it gets a lot of use, the Grapevine will be able to compute weighted average scores, where the weight of each rating depends on the influence score of the rater in the appropriate context.