Thread by @kritarthmittal

Meet Aleksandr Zhadan.

This 23-year-old developer from Russia used OpenAI to talk to 5,240 women on Tinder and automated dates with 100+ of them.

Now he is getting married to Karina, the best date found by his algorithm.

Here's the wildest Valentine's story you'll ever hear:🧵

It all started in 2021 when Zhadan's girlfriend dumped him after 2 years of dating.

He decided to find the perfect woman for himself and did what a regular 21-year-old would do as a rebound — swipe through Tinder like a freak.

He went on several dates but observed characteristic disadvantages like:

- they used to drink a lot
- had emotional swings
- there was stiffness

He felt repulsed and didn't want to waste his time in trivial chats.

So he programmed OpenAI’s GPT-2 to swipe on Tinder and talk to women on his behalf.

But GPT in 2021 was broken.

The program didn't generate the best replies and it kept forgetting the context of the conversation every 5 messages.

The dialogues quickly ended due to immature and at times, offensive responses.

This one-time GPT asked a girl to go on a date with him in the forest (this was first date) 😂

It also kept repeating the same questions which made Zhadan look like a dork.

Finally, ChatGPT was released and longer context memory along with custom tools were introduced.

Zhadan fine-tuned ChatGPT to make it understand him and talk more like him.

He also installed a filter using FlutterFlow so that it wouldn’t throw unnecessary messages in the chat, like a cringe story.

Then to find relevant girls, he added photo recognition in the web version of Tinder and trained it on his swipes on 4k profiles.

He filtered out all the profiles that were

- without a questionnaire
- had less than 2 photos
- does not have a job
- shows breasts in the photo
- had noisy photos

ChatGPT eventually found better girls and chatted for longer.

It swiped right 353 times, matched with 160, continued dialogue with 160, and finally, Zhadan met with 12 women.

The insane part? 50% MATCH RATE and 7.5% CONVERSION RATE for actual dates.

Out of 353 likes, he matched with 160. For context, the average match rate is 4% to 14%.

The model had become so perfect that it exactly understood what Zhadan wanted and who would want him back too.

Before going to the meetings, Zhadan asked ChatGPT to summarize and

1. Condense every 20 messages into 1
2. Build a profile of the girl

But there was still something wrong...

After the meetings, the girls usually behaved strangely or ignored Zhadan.

He figured that ChatGPT v1 was clumsy.

It would schedule multiple meetings at the same time and/or make promises that Zhadan wouldn't know of.

This one time, it scheduled a date and swore to the girl he would bring chocolate and flowers.

But when Zhadan arrived empty-handed, unaware of the promise, he seemed cheap.

He gradually updated his project as OpenAI released GPT-3, 3.5, and 4, and it slowly became a more thoughtful lover.

But Zhadan was not going to stop there.

He integrated Google Calendar and Telegram to manage his time and avoid conflicting meets.

Within TG itself, he created two databases of general and personal chats, and using FlutterFlow, he created trust levels for sharing personal information with certain people only.

And finally, the day came. GPT-4 was released – the best OpenAI model yet so far.

With this, Zhadan developed an algorithm that perfectly understood him and his wants.

He often asked ChatGPT for help while chatting with the person himself.

ChatGPT suggested Zhadan talk about childhood, parents, goals, etc to develop meaningful relationships.

With GPT-4 capabilities and Tinder Gold, Zhadan got 4,943 matches the next month.

UNBELIEVABLE.

Especially for a guy. For obvious reasons.

Putting the ethical conundrum aside, Zhadan was playing a perfect numbers game.

Zhadan was meeting with 4 girls at a time. It felt like he was a candidate for a dream job with several job offers.

But then, in December 2022, everything changed for Zhadan when he matched with Karina Vyalshakaeva.

The most charming girl was found.

Karina was very empathetic, cheerful, independent, and always on the move.

After a few in-person dates with Vyalshakaeva, Zhadan paused ChatGPT from speaking with other women on Tinder.

Zhadan was very happy and he didn't want this to end. But he feared he'd do something wrong.

So, for one last time, he turned to AI again. He made ChatGPT V3 to keep things going as expected.

V3 didn’t have to look for people, just maintain a dialogue. And now communication is not with thousands, but just with Karina.

Along with AI-generated texts, Zhadan also started talking to her in chat.

V3 as an observer would also advise on how to communicate better.

V3 would suggest Zhadan support Karina, avoid quarrels, and do activities together.

The relationship ventured offline, and the couple moved in together shortly after in 2023.

A few months into the relationship, Karina figured that she had been talking to a chatbot.

She was initially shocked. OBVIOUSLY.

But then she understood that the AI looked at her through the eyes of Zhadan and didn't sell lies.

She figured that Zhadan had put a lot of time into programming. And for her, it was okay if AI was used for something sensible.

She found her real-life connection wonderful and didn't walk out on Zhadan.

Wondering why? Artificial Intelligence. It did find the BEST fit for him.

“He spent a lot of time personalizing these prompts so, for me, it’s okay when used in a rational way.” - Karina in an interview

One day, ChatGPT V3 summarized the chat with Karina and recommended Zhadan to marry her.

At first, he thought that V3 was hallucinating (he never specified the goal of getting married).

But then he understood his train of thought.

Karina said that she wanted to go to someone’s wedding and ChatGPT thought that it would be better if it was her own.

Zhadan felt ecstatic. All his feelings rushed through him and he made up his mind.

I asked in a separate ChatGPT to prepare an action plan with several scenarios for a request like

“Offer me a plan so that a girl accepts a marriage proposal, taking into account her characteristics and chatting with her.”

He went almost according to the plan No. 3 and the good news is, the AI was right.

Zhadan proposed to Karina.

And she said YES.

What started out as a borderline unethical mass production of AI communications,

pretending to be him so he could play a numbers game and save time,

turned into a beautiful long-lasting story of love, compassion, and understanding.

The development of the project took ~120 hours and $1432 for the API.

Restaurant bills amounted to $2186.

Zhadan explains how if he had tried to do this all on his own, the same thing would've taken him 5+ years and ~$21,000.

Romance may never be the same.

Writing pickup lines for dating apps is tough and most of the time, cringe.

A fine-tuned model can come to everyone's rescue and we might see more of this soon.

At first glance, using ChatGPT to meet your spouse can sound cold and informal.

Yet, Zhadan and Karina formed a heartfelt relationship using AI, and there’s no certainty they would have found each other without it.

Zhadan is not offering this product to others at this time, mostly because it’s very expensive to operate—he spent thousands on this program.

The schematics of his model look insane today (Ps. It is in Russian)

Does this feel normal to you?

Would you go out on a date with someone who does not put in effort to understand you better

and rather let AI do all the heavy lifting?

I am not sure, I probably wouldn't.

But again, I got rizz, I can manage.

Most men out there probably are better off with a thoughtful AI model lol.

Anyway, if you enjoyed reading the thread, don't shy away and follow me @kritarthmittal for more!

Also, I am building @soshalsapp, an ecosystem for all the professionals and content creators out there.

Building a personal brand and posting viral content is going to get a whole lot easier.

Check it out here -

Ps. Feedback/suggestions are welcomed :)soshals.app/login