AI future is unavoidable. Building software for gods (me and my roomate) e/acc
Joined September 2023
- Tweets1.1K
- Following675
- Followers1K
- Likes1.3K
Dark retweeted
A few weeks ago we came across this tweet from @Morbiusdak about him creating an AI Dating bot that could Rate potential dates, Autoswipe, send a DM and hold a conversation.
So we partnered up to create Matchbox. We can currently take on only 100 users in our preorder phase
After talking/working with them for a while, I’ve decided to give this idea to @MatchboxGo. They reached out and made a startup for it.
They seem to know what they’re doing (they have a guy from amazon lol) and will be getting the software up on the cloud for it to run on browser💻
Kinda small team so they’re only taking like 100 people to use it at first. Go and support them if you like the idea.
Dark retweeted
Crazy things happening in the tech space, will be exciting to watch.
A new way of securing dates on dating apps using AI
After talking/working with them for a while, I’ve decided to give this idea to @MatchboxGo. They reached out and made a startup for it.
They seem to know what they’re doing (they have a guy from amazon lol) and will be getting the software up on the cloud for it to run on browser💻
Kinda small team so they’re only taking like 100 people to use it at first. Go and support them if you like the idea.
In all honesty I am a 6/10 white guy, 5’9” 190lbs. Been described as “ugly cute” pretty often. But still,
-1.3% swipes turn into matches?
-Less than half of them actually talk?
A literal RNG game, like a casino
🧵(2/3)
ALT This is INSANE! I bet the numbers are so different for women...
Some other stats:
-In >30% of the conversations the girl would not respond for over 36 hours. I never used to continue these convos because I don’t have time to wait on a woman like that, but a language model does. Funny enough, 5 of the 12 dates were with girls who went MIA like this so the lesson here is keep pursuing.
-Only 1.5% of the girls I matched with were “easy”. 3 of the 12 dates took less than 15 messages back and forth to set up. I think guys sometimes get on and want to fuck so bad they just think every conversation will play out like that where it’s really flirty for an hour and boom she wants you, but it looks like that rarely ever happens if you’re in my league ig.
-There is SO MUCH ghosting on the app it's ridiculous. I did not set it to send follow ups to girls that ghosted me, even my code is above that.
Here's a screenshot of one of the easy girls. LLMs have come such a long way, what do you know about "wife material"?😂
🧵(3/3)
After talking/working with them for a while, I’ve decided to give this idea to @MatchboxGo. They reached out and made a startup for it.
They seem to know what they’re doing (they have a guy from amazon lol) and will be getting the software up on the cloud for it to run on browser💻
Kinda small team so they’re only taking like 100 people to use it at first. Go and support them if you like the idea.
Dark retweeted
dafuq you say to me?
we don't have to wait until sunday, we can do a hackathon right now
who can write the smallest tetris program in haskell, blindfolded using only strings, no integers
also i invented bfloat thank you very much
just because im the pharma bro doesn't mean i can't program. the b in bfloat stands for bro. bro float.
This Post is from an account that no longer exists. Learn more
Making AI control a dating app was pretty easy
-Scrape website elements
-Extract profile pics, bio, etc
-Format the current conversation
-And give it all to a vision-language model to get a reply
And automating the swiping is literally just
-Download pics
-Sending to vision embedding model
-Give features to a CatBoost trained on my swiping preference
This matches my taste in women at ~84% accuracy^
In building this the real struggle was
- Getting the llm to talk like me
- And work with me in my daily life (ie don’t try to line up a date with someone when I'm busy/tired)
🧵(2/4)
To fix the first issue, I used a lot of in-context learning. I selected 4-7 good example conversations I've had and put them above the prompt as if the model had been speaking like me previously. This works surprisingly well on open-source models these days using datasets like Nous-Hermes, Capybara.
The second issue was more complex. Most chat-like language models have their system prompt where you tell them what to do. I put my own profile in there and told it "Be kind, be flirty etc". Then, I made a little LLM agent to chat with and made it responsible for changing the system prompt and alerting me to recent activity. I call it the Wingman.
I can tell it to:
-Try to get a date for the weekend, or one for every day of the week
-Only set up "drinks and a movie" dates
-"Be less lame" with the intros.
And I just got it working on my phone (GChat). God software is so sexy.
🧵(3/4)
So In the end, I have an AI that:
-Swipes according to my tastes✅
-Talks to my matches in my style✅
-Talks to me to match my preferences/availability✅
To be real, it definitely has its flaws. It will still say weird shit in 10% of conversations, but it's a numbers game now and one I can win.
Why did I build this? Because being on dating apps was annoying, wasting my time.
Shoot me🤷🏻
🧵(4/4)