ChatGPT Find Me A Wife on Tinder

🔷 Subscribe to get breakdowns of the most important developments in AI in your inbox every morning.

Fabulous tale of a Russian coder Aleksandr Zhadan using ChatGPT to find himself a wife on Tinder.

He split the process up into two steps:

• Finding a girl

• Chatting her up

When finding a girl on Tinder, he used a web scraper to get the images so that the AI could swipe right based on photos:

• minimum 2 photos

• in 2nd iteration, trained an image similarity model to find girls similar to the ones he already liked from 4,000 swipes

In the chatting up stage, in the Datebot V1 iteration:

• GPT-3 would always start the conversation

• The prompt was “You’re a guy talking to a girl for the first time. Your task, not immediately but to invite the girl on a date.

• GPT-3 was very bad at this, frequently forgetting the conversation

• Bot couldn’t access Telegram, so dropped half of potential dates

• Bot promised flowers or chocolate for dates.. and our tech bro showed up without them, and received complaints

Datebot V1 Results:

• 353 matched Tinder profiles

• 160 chatted (~45% of matches)

• 12 dates (7.5% of chats)

Not to be discouraged, our intrepid brother continued to build, and in the Datebot V2 iteration:

• Used GPT-4

• Each chatbot had a memory consisting of a background story for most simple dating questions

• Handoff from Tinderbot to Telegrambot for messaging

• Integrated Google Calendar for setting up dates

• Human in the loop verification for messages

Datebot V2 Results:

• GPT-4 hallucinations down to zero

• 4,886 matched Tinder profiles

• Countless dates: “so many it was scary”

The moment(s) of truth, time for date preparation:

• shower, shave, dress

• Ask ChatGPT to summarize conversation with candidate so far

He had many dates, and at times was juggling 4 girls until he found someone! Karina, oh Karina…Which then led to the build of a V3!

Datebot V3:

• to chat with just Karina

• Prompt “Keep on good terms with Karina, tell me if there is anything negative I need to pay attention to, or if I need to answer a question”

Side Project: Once he told Karina about his search (just numbers not technical details), she suggested he should use the remaining contacts, so was born a project:

• He found vacancies on Russian job portals willing to pay for employee candidate referrals

• Matched the conversations with the girls with the potential jobs

• Sold the contacts/relationships formed by GPT-4

• Arranged and got paid for 8 hires

Finally, one day:

• Karina chats about wanting to go to a friend’s wedding

• Datebot V3 tells our tech bro he should propose because Karina really wants to go to her own wedding!

• Proposal prompt: “Offer me a plan so that a girl accepts a marriage proposal, taking into account her characteristics, and chat with her.”

All in all, a somewhat unscrupulous, but heartwarming tale of boy meets girl, and girl and girl x 5,239 with the assistance of enough compute to power Soyuz to Mars and back. Bro’s real genius was in getting Karina to suggest the side project to offload his other prospects… or was that her genius?

Final Results:

• Dev time: 120 hours

• GPT API costs: $1,432

• Restaurant bills: 200 rubles

• Side project revenue: 526 rubles

• Estimated time and cost if he’d done it the manual way: 5 years, $150,000

• 5,239 Tinder matches → 1 wife

The Russian proposal video here.

Become a subscriber for daily breakdowns of what’s happening in the AI world: