Opt-in collaborative shared result ranking

carl

I sense a lot of... fear of the future/unknown/internet-populus floating around this conversation, and I think it's driving a lot of assumptions.

I agree that there are a lot of branding and bias management issues surrounding this concept. My opinion is that with some creativity and a well honed perspective, a lot of the perceived blockers can be used as advantages. If they are no longer treated as blockers, then progress can be made, rather than wait-and-seeing the feature to completion.

For Example: I don't assume that the terminus of this feature will not support multiple pools of site rankings that a user can create their own consortium from. A user may be able to select from a menu of sentiment pools, or be matched to a pool(s). I have no idea what this would look like, but i have several ideas surrounding the concept.

I loathe to mention it, but just think about reddit within this collaborative social search context. They have many many maaaaany self selected groups based around the core concept of up/down votes.

I will admit that I hand-wave away the activist element, and it may be bigger than I realize. My assumption is that features of this nature should be architected in such a way that is tamper-resistant. Given the kind-of non-single-kagi-user-pool thoughts I've been chewing on lately, I also suspect that a majority subset these 'activists' are just maybe the angry-pitchfork-rabble sentiment group. Ok .. welcome - knock yourselves out. here's how you socialize about where to throw pitchforks... so to speak. (or perhaps more accurately, here's what pitchfork-wielders up/down doot most )

As far as timing goes... I also have a different opinion. One could also say that the small tight knit nature of the current user pool lends itself to 'simple iterative' feature delivery at a time when growing-pains might be less impactful. Perhaps this feature will have refined to the point that it's ready to support the diversity of the internet, before said diversity shows up at the front gate. The small nature of the user group might be an advantage from this perspective.

Vlad

I wince a bit when the conversation shifts to managing negative impacts around politics and user bias. My personal strategy for socially charged features such as this, is to maintain awareness and rigorously architect a solution that bias-free enough that there's not really anything in the delivery pipeline to point at as evil or unfair.

examples: (i'm just exploring possible foibles, not suggesting implementation here)

1: I choose "users who love github" as a sentiment pool to enable. I review the upvote-only list and see that stack exchange and some others are also prioritized. Why isn't devdocs.io listed? dunno, maybe i should upvote it and get others checking it out!

2: I choose "users who love github" as a sentiment pool to enable. I review the upvote/dowvote list and become offended that devdocs.io is deprioritized. the nerve of these kagi users, would do such a thing. I'm upvoting it.

3: I can't wrap my head around what a downvote-only capability would look like - insert downvote-only scenario here. why is devdocs.io deprioritized? this is ludicrous. it's just a reference agregator. I might just go chew someone out on kagifeedback. this is unacceptable.

I chose to use the "Users who also love" sort of paradigm very intentionally earlier in the thread. It's hard to target percieved political bias when the dimensions of grouping have nothing to do with politics. Sure there would be biases within user's shared sentiments, but they might be bucketized in a way that diminishes that impact perceptually and functionally.

Being able to create shareable and modular lists of adjustments to your rankings would be nice for crowdsourcing the labor of curating results.

This builds off my "bug" report and this feature suggestion about marking sites as AI generated together with my experience using Bluesky's Moderation Lists

Note: I know there's Lenses which can be shared, but their functionality is really meant for, as the name implies, very specific and repeated search situations.

Ranking Lists would be groups of adjustments that could be subscribed to and merged into your existing rankings (where your "global" adjustments take priority). This way if it would be possible for people to say highlight a collection of high quality blogs or provide a good starter kit for addressing SEO mills about a topic they follow (programming is obvious, but it easily applies to gardening, cooking, and hobbies generally).

When making an adjustment on the popover there would be a way of adding it to a list (graphic design is, in fact, not a huge passion)

A ranking adjustment could be on multiple lists, but (and I recognize this might be controversial to some people) the adjustment itself would be global to your account. If you're deeming a website useful or not in general it's context shouldn't matter. That's where Lenses would come into play.

Because of that I could see the current Personalized Results settings page continuing to exist but having a way of seeing the lists a domain is on (maybe the same list selection interface as the popover), plus an additional detail page for managing/viewing lists individually. More or less replicating what Bluesky has going on.

Some design thoughts:

• There would need to be a visualization indicating if a modification you've made overwrites something on a list or if a domain was added by a list and not by you
• Conflicts between lists (since it's not just "here's a list, take one specific action" as with Bluesky) would need to be addressed
• It may make sense to limit lists to Raise and Lower with no Pins or Blocks or making accepting those an option, similar to how Bluesky has Mute or Block subscription types

I'm sure there's a world of technical reasons this would be difficult to implement in the backend, but I think it would be a great addition to the platform!

P.S. the Personal Results settings page uses a Select for the selected adjustment, but the Domain Leaderboard lets you choose using a Radio Button Group

This is a pivotal feature and Bluesky is showing/leading the way. Their approach is a good proof that users are sensible to ranking algorithms & will use them.
This helps Kagi provide more value & personnalisation without associating itself with an idea or another.
Getting the ball rolling on "Custom Lenses SDK" early on will add another dimension to Kagi, I guess a good part of the userbase is tech literate but building a strong dev community & open source ranking algorithms must take a little time !

It's interesting to see the comparison with the Bluesky collaborative ranking tools. This is a critical feature of decentralized social media: block lists, promote lists, etc. Same with adblockers... domain reputation is everything. uBlock Origin is only as good as its block lists. It would be amazing if Kagi to published these kinds of lists.

It would be something like a Private Tracker: it's a database curated by and for a defined, trusted population. In this case, people who care enough about search quality to pay for Kagi. Kagi would need to be very thoughtful about how to receive trusted feedback from users without reducing privacy. Privacy Pass might be a valuable foundation for this.

Lets see if I can state this as clear user stories...

As a Kagi Subscriber, I want Kagi's high-quality signals about publisher quality & relevance available on other platforms. When Kagi thinks a domain is used for spam or other malfeasance, I want my adblocker to block that domain. When Kagi thinks a user on a decentralized social media platform is fake account, spammer, or otherwise fraudulent, I want that user to be automatically blocked from my feeds on that platform.

As a Kagi Subscriber and decentralized social media user, I am willing to share my personal moderation decisions pseudonymously with Kagi in return for access to blocker feeds/moderation lists. I don't want Kagi to build a database of my social media behavior explicitly tied to my account, but I'm comfortable sending a stream of block/unfollow/report events tied to a stable ID that is not otherwise linked to my Kagi account. I expect that Kagi will use this identifying information exclusively to prevent abuse of the reporting service.

As a Kagi Subscriber, I want the ability to search social media at scale, including the long tail of decentralized social media instances. I expect these searches to be ranked based on high-quality signals. Users who share links to low-reputation domains or are frequently blocked by other Kagi users should be downranked.