Yet Another Dating App (some ideas)

Now before you raise your pitchforks shouting. “there’s too many apps”, let me just say that it doesn’t hurt to keep trying. We can totally have dating app #38.

Background

Motivations

Dating is the one thing most people won’t admit to struggling with.

The internet has fundamentally altered how we interact with one another, and while we are talking to each other a lot more, it feels like we are not having nearly enough micromarriages.

There is a lot of discourse online on dissatisfaction with dating apps, and most people agree that it sucks. ¹ There exists a rather small subset of users who enjoy using the currently available popular apps, despite the fact that there will always be a market for this. Believe it or not, there will always be people looking for love. It’s why we’re here in the first place.

Anyway, I came across this discussion on the SSC sub and it gave me some insight into what it would take to build Yet Another Dating App.

A bunch of different dating apps have emerged over the last few years with various, different value propositions (The dating app to be deleted, the dating app for [insert your in-group]). Some of them reportedly work well for a while, get gamed and then kind of fade off into obsolescence. The UX is mostly the same (a couple pictures, a simple bio, Elo ranking, swipe-based matching system), or something to that tune.

The fundamental problems with pretty much all dating apps are the same though: the business model. At some point, all dating apps gradient descent towards extorting their users for a “better” experience. Many users still report dissatisfaction even once they have subscribed to a paid tier²

Tangent

I’m starting to believe that maybe this whole dating market fiasco is not just a result of one big factor (a la dating apps trading user satisfaction for profits) but actually an amalgamation of many different highly consequential factors. For instance, there is some chance that this is all some big PSYOP campaign by a foreign enemy (think China, Russia). You don’t really hear dating struggle stories from this part of the world. (Maybe cause censorship?) One story I do hear often from the East (Japan) is that there are too many single women who are willing and able to start families but a lot of men are retreating to singlehood.

Epistemic Status on why you’re not getting laid

I understand that some of these are a little outlandish

I could give a million more reasons as to why dating right now sucks but this is a way better, more in-depth analysis on probably why you’re not getting laid.

Yet Another Dating App (YADA)

Here are some ideas I had for a new dating app design.

Features

For the sake of simplicity, I’ll try to keep everything heteronormative

• Strict 1:1 gender ratio
  There can only be one guy for every one woman and everything is on a queue-based system. Why? Many (both men and women) complain that dating apps are a cesspool because it’s a sausage fest. The queue-based system works by having everyone join some sort of a waitlist and when there is room for an extra person, they get an invite to the app.

A 1:1 partner ratio creates a bipartite graph type problem. We also want to aim for a solution that is along the lines of a Gale Shapely (this is allegedly what is used in Hinge) algorithmic solution rather than a primitive matching system (like MCBM). Our goal is to maximize happy pairings, not just pairings. Of course, one challenge with this is that there’s no easy way to create strict rankings of potential dates for each user. ³

Maybe we can have some sort of Looksmatcher AI do this for us (make a top 5 list for each man and woman) and then model it purely as a Gale Shapley problem.

• If you don’t match with anyone in 2 weeks (or some arbitrary number that should be determined by sound statistics), you get kicked out of the app

• In-built call feature
  Have a minimal call feature that lets users talk to each other without having to share their phone number. You might as well throw in a FaceTime feature too.

• Recreation of the charm of early OkCupid: a long, personality quiz that determines your likes and dislikes. You then only get matched with other users who have compatible interests

“It’s 2023, there’s no reason I shouldn’t be able to search for other people into “horror movies” or whatever interest in my area. This shouldn’t be too hard to do, but afaik no dating app does it anymore because it makes matchmaking too efficient (ergo, users don’t spend as much time on the app)”

“I think the secret sauce of OK cupid was that it required everyone to start by answering their personality quiz, which was gendered. And if you were male, and you answered something like “not much dating experience,” it labelled you with “manchild” and a really insulting description that not-so-subtly asked you to not use their site. It also stuck their on your profile, unremovable and unchangeable unless you deleted your account and started over. So it filtered out a lot of undesirable males, and made it really easy for women to avoid them.”

More on this in a bit…

• A non-profit governance structure

“The basic issue to solve, in my view, is the incentives. Your app will invariably gradient-descent towards the Tinder model by default, because you don’t have a way to be rewarded for creating happy relationships.”

• A computer vision model can rank people’s attractiveness and play matchmaker.

" One area where I think most matching sites fail is that they don’t capture attractiveness, and end up matching people who are similar on paper but where one is way outside the other’s league in terms of attractiveness"

“Piggybacking off of this, AI has gotten way better overall, but to my knowledge nobody has trained a modern transformer/deep learning model to play matchmaker. Creating a label for “how good of a match are these two people” to train on is tricky, but as a proxy I would look at how many messages two people send back and forth to use as a training label. For features I would use the ones extracted from attractiveness model above, and a LLM to parse through their profile text. Train a model to predict how many messages two people will end up sending back and forth. Match people in such a way that maximizes expected messages back and forth. This has a dual benefit of incentivizing people to talk a lot on the app, because if you’re boring you’ll get matched with other boring people.”

The obvious problem here is: where in the world could you possibly get this kind of data? I guess people who truly beleive in the mission would volunteer their own data? Hmm

Cloning the best of OkCupid

This seems to be a widely suggested approach. It seems like most people who used the app in its golden years (circa 2010) have a positive review of it. Here are some of the main features from that period:

• Searchable profiles (see everyone in your area, not a secretary problem system where you get one profile to accept or reject at a time)

• A “match percentage” that can be gamed as a rough IQ test (i.e., you could answer and mark as important questions like “2x+8=13, what is X?”) or used to filter on important values. Also searchable. Also, make it so people can search for questions, change their answers, or delete questions they’ve previously answered.

• Free messages that are guaranteed to go to the recipients inbox (as opposed to the modern, we’ll talk if we mutually swipe). One of the justifications for disabling this feature was that women would get overwhelmed by messages or harassed by crude come-ons, but I think these are rationalizations not reasons. It would make much more sense to ban harassing users and/or implement a “limit to one initial message” rule than to move to Tinder-style swiping (no messages until mutual swipe).

  • *This probably already exists in current apps but one thing to look into is whether a large majority of harassment texts come from a small number of people. If yes, we could easily ban anyone who misbehaves, and set up some sort of ML model that detects banned faces.

Current unconventional offerings

There exists a couple simple, fringe services like Date Me Docs that take a more unconventional approach to dating apps. It expectedly has a fairly small user base. I’m not entirely sure what its success rate is. Cuties (F.K.A. twitter dating app) is another one of these services that actually launched on X If you use any one of these services, let me know your success story

Decentralized dating app?

Governance-wise, it seems as though the biggest problem with dating apps is that there is a centralized entity that is trying to milk as much profit as possible, which tanks the experience for most users by:

• giving them a limited number of likes/swipes
• forcing them to shell out more money for more visibility
• making bots to feign engagement (allegedly)

So why not make a decentralized dating app? No one party owns the platform, so no one gets to profit off of it. Given the recent success of various decentralized social apps (Bluesky, Mastodon, PixelCam), it seems like this is a decent model.

Criticisms of a federated approach

• The tiny faction of people who know about decentralized applications is largely dominated by: male, software developers. Having a subset of that (single male devs) is probably not helping the value prop.

“With the best will in the world, even if I was interested in dating, I’m not sure I want to have my dating pool just be a bunch of Linux-using software developers.”

You can argue that this lot are the early adopters, and regular people will join in on the fun soon after.

• Naturally, people will not want their social media presence to be “linked” to their dating app account; which I believe is automatic with protocols like ActivityPub.

• Safety: How do you enforce safety policies (e.g. to fight harassment, catfishing) without some centralized party? Who does the moderation? Server hosts? That’s not gonna fly with most people.

• It’s too complicated. People are just used to signing up and logging in to start using some app. They couldn’t care less about what “instance” they’re on, and the overwhelming majority wouldn’t care as much to “start” their own instance.

Maybe before launch, we could pool instances and have them all assigned to some certain archetype. When people sign up, we could automatically assign them some instance based on their algo-determined archetype and match them appropriately. As you can imagine, this will be gamed mercilessly once people realize which instance all the hot girls are.

• It’s too expensive Say we do manage to get a decent technical solution going… how do we actually get more people to sign up. Dating apps are social apps after all, and this is the law of network effects. No one wants to use an app no one else is using. Currently the CPA (cost per acquisition/download) of a modern dating app is pretty high (~$10-$20 per user for a new app and ~$3-$10 per user) [^5] Most of these costs come from advertising for brand awareness, which is how people know it’s legit.

Anyways…

I’m a little skeptical about whether this is the kind of problem that can be solved with more technology. Like, can we solve this deep biological bug feature by just throwing more compute at it?

Some part of me is convinced this may not be the right approach. I keep thinking about how “oldheads” swear it was easier to date before the smartphone era. Maybe there’s some truth there. Maybe the friction they’re nostalgic for (the lack of infinite options) wasn’t a flaw, but a filter. When your dating pool was limited to people you crossed paths with organically (friends of friends, coworkers, the regulars at your coffee shop) there was a kind of baked-in curation. The algorithm that is life’s randomness did the work: proximity, shared routines, overlapping social circles. You didn’t need AI to predict chemistry because chemistry revealed itself—awkwardly, slowly, in stolen glances and bad puns.

Now? Infinite swiping turns romance into a hypermarket of hypotheticals. Paradoxically, more options = less clarity. Algorithms prioritize “compatibility” based on hobbies or politics, but they can’t quantify the unspoken spark that flickers when two strangers laugh at the same stupid thing. Worse, they ignore the value of constraints: scarcity forces investment. When you can’t ghost someone and instantly match with 50 others, you’re incentivized to actually try.

Maybe love isn’t a problem to optimize. Maybe it’s a thing that thrives in the cracks of imperfection, in the unplanned moments we’re too busy engineering away. I wonder if we’re solving the wrong puzzle—or worse, inventing new ones.

Other References

Can a dating app that doesn’t suck be built? Let’s brainstorm
Look at the real world, the reason nobody is building dating apps is that user acquisition is expensive
Dating Roundup #1 by Zvi
Maximum Cardinality Matching