Aashiq banaya AI ne (AI made me a lover) || AI generates Love Quotes
So, Valentine day was near and after posting the last video “AI shuts my GF mouth”, it was very important to impress my girl. But I’m not that good with romance or love lines. I seek out my friends but they are as noob as I was. So, I thought of developing an AI that can generate unique love quotes for me that I can share with my girl. This video is my journey, explained in a sarcastic and informative way via my storytelling skills.
Feel free to checkout my last article — “AI shuts my GF mouth”
This is a NLP based problem, or a generative task to be exact. Transformers are dominating the NLP domains now a days. I did some R&D about transformers and then I found 2 well known transformer model i.e., BERT and GPT. Since, it is a generative task, GPT suits our problem statement. I used a dataset “Quotes 500k”. It contains 5 Lakhs of quotations from different genre. First, I clustered out the labels related to love and then did some data preprocessing. After that, I trained a GPT model using our custom dataset and results were amazing.
The AI did produce some that would pass for — unique love quotes
“The love of a woman in my heart is what keeps me strong when she is not around.”
“Sometimes you just want to be loved and you just want to be loved and then you have to fight for your love.”
“I do not know if i could be in love with someone as beautiful as you”
“Love is the highest power in the universe”
“I have had the pleasure of seeing you, and i am glad to have you back.”
These are few of the quotes generated by our AI. Seriously! Our AI model turns out to be quite good unexpectedly.
Sometimes, it gives meaning less quotes like:
“Love is a kind of love, and you cannot live without it.”
“If it hurts, it does not matter. it is not love. love is not love”
“I love the way you smile when you smile, and the way you laugh when you laugh. i love the way you smile and laugh when you smile. i love the way you smile and laugh. i love the way you smile and laugh and laugh. i love the way you laugh and laugh and laugh and laugh and laugh and laugh.”
But they are very few cases.
Each quote in the dataset (Quotes 500K) is associated with multiple tags.
Though using regular expressions, we segmented all the quotes with the tag containing text related to love, but due to the the fact that each quote contains multiple tags from different genre, our AI learnt some unique patterns that it was not intended to learn. Some of them are given below:
- Great Motivator
“If you’re feeling hurt, give yourself time to heal.”
“Do not let yourself be discouraged.”
“I am the light of my life. i am the light that always shines brightly”
“It is as simple as that : when you say you are sorry, no matter what your response is, you feel it. when you say you are happy, no matter what you say, you are living a beautiful life.”
- Depressing
“Love is like a dream. it is like a dream that can never be fulfilled.”
“I love you so much i am afraid i will never see you again.”
“Sometimes we get what we need. sometimes it comes too late.”
- Eh, Idk
“If you never really loved anyone, you were never likely to get along well. you were a pretty fucking pathetic little ass to me because you were the only one who really had the heart to live up to your own ridiculous expectations.”
“I would rather die than live in a world where i am a prisoner of the law.”
“I do not want to live in a society with someone who is not happy with the way i am”
“I have known someone who is not as good as i am, but i am willing to pay.”
But in the end, I’m happy with the results.
The video shared here consists of my experiences while training a generative model using GPT and the results that i got were pretty amazing as i’ve also shared here. Feel free to check out my experience using the given link. Hope we can relate with the experience of training a large model or else, if you are new in it, i hope it’ll bring some value in your life.