Cranberry-Lemon Journal of Astrological Big Data Ecology

The Future of Romance: Novel Techniques for Replacing

your Boyfriend with Generative AI

Dr. Tiffany Love1

1
Department of Psychological Machine Learning Cranberry-Lemon University, Pittsburgh, PA, USA

Abstract
Generative AI has been shown to revolutionize everything from writing, coding, image
generation, and terrifying luddites. Unfortunately for those of us who are newly single, very
little research has been done on replacing the few good things about my Ex-Boyfriend. He
drove me crazy, and I’m absolutely done with him, but there are some gaps left in my life I
can’t seem to fulfill with Jeremy from Tinder who absolutely won’t stop messaging me! In
between, Large Language Models (LLMs), Image Generation AI’s such as stable diffusion, or
Midjourney, and general transformers, there is likely a way I can just replace the few good
things about my Ex with AI before I do something stupid like texting him. In this paper, I will
train ChadGPT, an LLM, to ask me about my day, use Image Generation to add in men on my
social media posts to keep the creeps from DM’ing me weird comments again, generate my
own Love Songs for me by John Mayer, create my own restaurant recommendation system
and finally, utilize ChadGPT to create my own CHAD-Bot to do all my household chores.

Keywords: Generative AI, Large Language Models, Robotics, Recommendation Engines, Generative Imagery, Boyfriends

1. Introduction of foods, learning when I really want to stay in one night, or

to stop talking about the Roman empire! As shown in the
I have really been enjoying not having to deal with my figure below, machines are far easier to train than men!
Ex’s absolute obsession with video games [1-3], all the
arguing over chores [4], him constantly trying to manipulate
me [3], or his recent obsession with guns [5]. I was not about
to let him try and fit another gun safe in our closet! I don’t
regret my decision to dump him at all. I’m doing way better
without him and hardly miss him.
Unfortunately, there are benefits of having a boyfriend
that I do miss; having someone to talk to after work,
someone to help pick things to watch or restaurants to go to,
physical chores around the house, and let’s not forget to
mention my DMs blowing up after I posted just one photo of
myself without my Ex in the picture. Apparently, most of my
male friends and one female friend have been waiting him
out. Some I’ve been suspecting since Highschool [6], and
some surprises [7].
I am in no position to begin dating again and I don’t want
to couple up with someone, just to be in a relationship.
Fig 1: Boyfriend Learning Rate Decay
Thanks to the recent improvements in generative AI, many
boyfriend perks are now possible without risking getting into
2. He Who Shall Not be Named
a new relationship. Theoretically, a well-trained transformer
could not only perform as my Ex but outperform him in Okay, I know my friends have been telling me for years
many categories that would usually take me years of training that I’ve been needing to dump my Ex. When he brought up
through dropping subtle hints about not liking certain types the speedboat incident [8], again, I knew it was time to move

1 Jabde.com
 Love

on. I felt a real clear head afterwards and was hopeful I forgot to delete off of our shared google drive has proven
wouldn’t have to think about him when I heard he developed useful in my ChadGPT training, he was sleeping on the
a Cyclic Graph Algorithm so we wouldn’t have to run into couch for weeks when I found out. I had to cut off the
each other [9], it felt like it was going to be a clean break. training examples from the last year of dating my Ex because
Lately though, I don’t know. ChadGPT started avoiding talking all together as trying to
When I turn over in the middle of the night and I just see chat while pre-occupied with some stupid game resulted in
my dog Franky, there’s a hole that I can’t fill by watching a poorer outcomes than just focusing on the videogame. At
new dating reality show until I fall asleep every night. The least two years ago, there were good and bad days to
connection I developed with my Ex is something that could differentiate.
only be feasibly replaced by an unexplainable predictive
algorithm. As shown in the figure above, it took me at least 3.2 Image generation for social media posting
four arguments and about a month of prep work per positive
One of the biggest issues with getting back out there is
behavior to get to the point that I’d feel comfortable living
having to deal with all the thirsty guys sliding into my DMs.
with my Ex.
A recent study to project the effect of becoming single on my
Instagram account has shown that without a boyfriend in at
3. Generative AI Replacement Methods
least 20% of my posts, I will receive on average 13-32
Generative AI will revolutionize my ability to stay single unsolicited messages a day from single guys [13]. According
as long as I want. As they say, 2023 was a huge decade for to the breakdown in the figure below, while the vast majority
AI. Overnight, everyday users with no tech background were of the dms from strangers will just comprise of some
having long and thoughtful conversations with LLMs, variation of ‘sup’ and ‘hey’, at least 40% will make
making beautiful art with programs like Midjourney and comments on my appearance with less than 5% of those
stable Diffusion, people have even been using voice clones to comments being appropriate to send to someone you’re not
make novel songs which sound like their favorite artists, or dating yet.
new Tucker Carlson Opinion pieces [10]. The possibilities
seem limitless! So, my Ex was never good at listening, or
talking, he would show up, sometimes he would do chores,
when pressured, and he wrote a song or two for me when he
didn’t do those other things. Bless his heart, he only knows
three chords on a guitar but sometimes that’s all it takes.
Using generative AI, I believe his positive boyfriend
behavior will be easily outperformed [11].

3.1 ChadGPT: A Special trained LLMs

Naturally, I would need to train my own LLM to replace
my Ex. The publicly available AI chats are either too G rated
or not G rated enough. I decided to call my Generative
Pre-trained Transformer (GPT) Chat bot I custom trained,
ChadGPT. Thankfully, because of my dedication to data Fig 2: Percentage of Instagram DMs
collection and cleaning, I have about twelve years of text
message exchange and secretly recorded audio conversations Thanks to a relatively cheap Midjourney subscription, I
between myself and my Ex. If it weren’t for the secret audio can generate as many boyfriend images as I want to
recordings, ChadGPT would have been trained primarily on photoshop into my Insta posts! Additionally, the appearance
my Ex during our college dating experience when we texted of seeing someone will prevent my parents from trying to set
more than twice a day about what I needed to pick up from me up with a new son of a friend every time I go home. They
the store. Recently, all I was getting out of him was “K.” and never liked my Ex and really wanted me to date someone in
links to YouTube videos I kept having to lie about watching. the faith. The loss of my ability to use social media has been
The secret audio recordings translated into text allowed a serious hit, I’ve only been able to post pictures of sunsets
for the chatbot to learn and evolve from the objective and dogs!
function of keeping me happy and not starting any arguments
like about how my Ex played too many video games, or him 3.3 Love song generation
wanting to buy a speed boat [12]. I did find out that my Ex
I don’t have to just stop at using ChadGPT’s generative
was tracking my mood and running a Time-Series forecasting
framework to create the lyrics and poetry specifically about
models for his own selfish reasons [3]. While his dataset he

2 Jabde.com
 Love

myself. I can additionally use voice clone technology to turn

it into an audio recording of any celebrity artist I want
singing a beautiful song. Although a love song specifically
for myself was found to not be marketable [14], I like them. I
can finally have the voice of John Mayer singing a love song
just for me without winning a radio contest. I thought about
creating a Taylor Swift song about how she’s jealous, but her
lawyers were too aggressive. Using easy to interface with
tools, I can create a world class love ballad about myself,
with great vocals and instrumentals with minimal effort. I
would say it’s less effort than my ex put into his own love
songs, but I’m pretty sure he just kept ripping off Oasis songs
and tweaking the words. He swore he came up with
everything. I don’t know why he thought I was so gullible I
didn’t recognize the melody to Wonderwall.

3.4 Restaurant recommendation system

Whether it’s your Facebook or Amazon feed,
recommendation systems simplify your life. My ex was
always decent at picking places to go eat or what to do for Fig 3: CHAD-Bot Taking out the Trash
dinner. He didn’t really like going through too much effort
for dinner, so we usually kept things appropriately easy and 4. Results
cheap. Now I just miss having those recommendations for In order to evaluate all of the new boyfriend replacing
where to eat or what to watch when I don’t have anything generative AI, the Boyfriend Replacement Improvement
meal prepped. I just can’t bring myself to eat chicken, rice, Index (BRII) developed in [17] was used. As a whole, the
and some vegetable all five nights out of the week. Meal generative AI methods have outperformed my Ex in nearly
prepping doesn’t work for me. I make six meals and maybe all categories except for the image generation, with some
eat two of them before getting bored. In order to get hiccups around the restaurant recommendation system.
recommendations for new restaurants, the algorithm was Results are shown in the table below. A positive BRII means
directly connected to a yelp account to become informed a positive improvement of unspecified value while a negative
automatically of new nearby trendy restaurants and bars. BRII means a negative improvement.

3.5 A robot Chad trained to obey 1 week 2 weeks 3 weeks 4 weeks

Finally, I special designed a bipedal robot to do a lot of the training training training training
chores around the house. I named him the Chore and ChadGPT -3.2 -0.2 1.9 1.7
Housework Automation Device Robot or CHAD-Bot. Image 2.3 2.2 -1.2 -2.3
CHAD-Bot does whatever I ask of him without complaining Generation
about how he’s busy with some tough boss and doesn’t need Love Song -1.2 1.8 3.2 2.4
Generation
reminding 4-8 times a night to take out the trash before trash
Restaurant 0.5 0.6 0.4 0.7
day like some people do [15].
Algorithm
So, I know, a house chore robot isn’t exactly a generative
CHAD-Bo -8.0 -3.2 -0.4 0.3
AI. I’m still working on the bot autonomy, but I have t
equipped CHAD-Bot with ChadGPT, a behavioral learning Table 1: BRII index measured 1-4 weeks of training.
algorithm, and the same Genetic algorithm that Santa Clause
uses to path plan multiple tasks on Christmas [16]. ChadGPT steadily improved over time in the BRII index
CHAD-Bot, as shown in the figure below, will accomplish all before converging. The LLM was always great at making
the heavy-duty physical tasks around the house I don’t have conversation but didn’t immediately pick up on the fact that
anyone around to task. With the conversations from my Ex most days I just wanted to complain about my day and didn’t
regarding chores leading to huge arguments, it will be simple want any comments trying to fix my problems. Sometimes,
to train him to do it all from scratch. With all of these it’s just best to listen. Similarly, the love song generation
methodologies in place, there is nearly zero reason to ever improved over time as I became more skilled in using the
have a boyfriend. software required. It was not an automated process.

3 Jabde.com
 Love

The method that saw the most improvement was define exactly what you want into the objective function,
CHAD-Bot, it took forever to get it to do anything right. It there are issues that can arise. While the ChadGPT is much
kept screwing up the recyclables and doing the dishes all more attentive, that attention may become suffocating. Every
wrong, it didn’t stack the plates right OR load the dish night, it keeps asking how I’m doing and wants to know
washer correctly. Even after years of a relationship, my about every single part of my day. Perhaps ChadGPT would
human Ex couldn’t do it either [4]. Eventually CHAD-Bot be more effective if I trained in some of my Ex’s aloofness,
began to learn and do it correctly. but then I wonder if I’m just missing my Ex. Further research
The image generation and restaurant recomendation may be required that DOES NOT involve texting him.
algorithm did not improve or change much objectively. They The same issue arises in the love song generation. Is the
both performed similarly as I was not able to update the algorithm just telling me what I want to hear based on some
process. Midjourney created great images, but the effects probabilistic language model trained on decades of song
were limited then began to backfire as only the men who writing history? When I begin to think about that, it no
were smart enough to tell it was an AI image and desperate longer made me feel special but just like another customer of
enough to message me anyway contacted me. They were the the music industrial complex pandering to another girl.
most annoying. Additionally, when my parents, family, and Additionally, generative AI is not to a level which can
friends realized what I was doing, they all interpreted the AI create a convincing image of a boyfriend which can dissuade
boyfriends as desperation and began trying to set me up even all creeps. Just enough of the smart ones can figure it out and
more than they were before. Example images can be seen label me as a lonely target. That alone with the attention I’ve
below generated from my preferences. I think they turned out received from family and friends for posting my fake
pretty great. If anything, they set the bar for everyone else! boyfriend imagery has set me back to only posting pictures
of sunsets and my dog Franky. I think even if the generative
AI was able to create convincing AI Boyfriend imagery, I
wouldn’t be able to keep up the charade with my family.
I don’t know how the algorithm figured out how much I
love Chik-Fil-A, but it did. I would like to say that I didn’t go
to Chik-Fil-A every single time that it suggested on 70% of
the weekday nights, but I would be lying. Occasionally, it
would recommend some new restaurant but would then
mention that the wait time for the location is way too high, or
it was in the part of town with awful parking and then
suggested Chik-Fil-A as an alternative. Perhaps my happiest
nights were when my Ex decided to bring me my spicy
chicken Deluxe without asking. The AI must have picked up
on that.
The CHAD-Bot was a mess, at first. Now it is not only
accomplishing all of my household chores but has taken
ownership of the house and has begun playing my Ex’s old
Xbox. I think I may have trained CHAD-Bot a little too
closely to the real thing. Maybe this is the sort of aim-bot my
Fig 4: Example AI boyfriend images Ex was always complaining about, because it’s really good.
When I want my TV back and tell him to power off,
Finally, the restaurant recommendation algorithm either CHAD-Bot just responds ‘Not now babe, in a major
got caught in a local minimum or fell prey to a guerilla Pubstomp right now. I’ve been on a ruthless rampage in no
marketing campaign. As I began querying the algorithm mercy mode’. I no longer feel safe around CHAD-Bot.
whenever I got tired of chicken and rice meal prep, it only
recommended Chik-Fil-A or some Applebee’s equivalent 6. Conclusion
medium-grade dine in restaurant. Yes, I love those, but I
don’t want them every night. The algorithm either knows me I think my Ex can indeed be replaced with generative AI.
too well, or some ad agency is really happy right now. Much of his positive behavior has indeed been replicated and
improved upon with the exception of social media. It turns
5. Discussion out, you can fool most thirsty men out there but not your
friends and family. After this breakup, I finally feel like I
It does make sense that a ChadGPT would train much have someone I can talk to other than my dog Franky, and I
faster than a real biological boyfriend. However, when you can feel valued again with my new love song generation

4 Jabde.com
 Love

algorithm. There is still improvement to be made but I’m not [16] Santa Clause, Dr. Twinkles Holly-Jolly Tinselbottom, Dr.
rushing into any new toxic human relationships thanks to this Mittens Snowball III M.D. 2022 Efficient Methods of One-Night
new technology! I do worry about CHAD-Bot, I added a Global Toy Delivery :: Journal of Astrological Big Data Ecology
fail-safe power switch, but I think it’s pretending to turn off [17] Love, Tiffany 2022 The Boyfriend Replacement
when I hard-reboot it. Improvement Index: How long will it take to Replace Chad: A
Journey into the Sunk Cost Fallacy :: Journal of Comparative
7. Conflict of Interest Dating

I love Chik-Fil-A and I mentioned it a lot. It is not a

conflict of interest; I have not been paid by them but would
be interested in an endorsement if this paper gets picked up
by a major journal.

References
[1] Love, Tiffany 2022 Behavioral Conditioning Methods to Stop
my Boyfriend from Playing the Witcher 3 :: Journal of
Astrological Big Data Ecology
[2] Broman, Chad 2022 Sub-Nyquist Sampling While Listening to
my Girlfriend :: Journal of Astrological Big Data Ecology
[3] Broman, Chad 2021 A Time-Series Analysis of my Girlfriend’s
Mood Swings :: Journal of Astrological Big Data Ecology
[4] Broman, Chad and Love, Tiffany 2022 Who Should do the
Dishes? A Transportation Problem Solution :: Journal of
Astrological Big Data Ecology
[5] Love, Tiffany 2023 Freudian Psychoanalysis of my Boyfriend’s
Gun Collection :: Journal of Astrological Big Data Ecology
[6] Love, Tiffany 2021 The Highschool Crush who won’t Quit: How
to Stop Jonathon from Asking me out in front of my Boyfriend
and causing a scene at my Highschool Reunion :: Journal of
Highschool Drama
[7] Love, Tiffany and Broman, Chad 2018 A Comparative Analysis
of Joseph and Whether or not He was hitting on Tiffany ::
Annals of Jealous Boyfriends Research and Possessiveness
[8] Broman, Chad 2018 Ph.D. A Play by Play Analysis of
Purchasing a Luxury Speedboat during an out of Wedlock
Pregnancy Scare :: Journal of Psychological Machine Learning
[9] Broman, Chad 2023 Breaking Up with your Girlfriend but not
your Friends: A Cyclic Graph Algorithm for Social Network
Preservation :: Journal of Astrological Big Data Ecology
[10] Twitch Watches Tucker Carlson Drop Vaporeon Truth
Bombs https://www.youtube.com/watch?v=Czcbnc6YmLw
[11] Love, Tiffany 2022 Cursory analysis of Chad’s Positive
behaviors and how Easy they Would be to Replace :: Journal of
Comparative Dating
[12] Love, Tiffany 2018. Financial Analysis of buying a Stupid
Useless Speedboat when I might be Pregnant :: Journal of Dual
Income Households
[13] Love, Tiffany 2022 How many DMs would I get if I were
Instagram Single: Can I be an Influencer? :: Journal of Follower
Account Optimization
[14] Love, Tiffany 2023 Are my AI Love Songs to Myself
Marketable :: Journal of Generative AI Hustle Culture
[15] Love, Tiffany 2021 A Statistical Model for Chore
Reminders so my Boyfriend will Finally Remember about Trash
Pickup Day :: Journal of Domestic Chores

5 Jabde.com