My Latest Crazy AI Project

0:00I'm an AI engineer and I successfully created one of the world's first AI

0:04trading platforms and today I'm creating a new AI agent that will make me

0:09go viral on TikTok

0:11If you're a software engineer and you're struggling to find these big tech projects

0:15that actually make you stand out amongst the crowd then stop and save this

0:20video So for context I've been posting on TikTok Instagram

0:24YouTube and X for a little over three years and I've also written a

0:29bunch of articles on my medium

0:32I'm using all of that data the raw videos the images within the articles

0:37the converting the gifs using all of that data creating embeddings for all that

0:43data and I'm going to train a model to try to predict what videos

0:48are going viral based on their video or their written article embeddings

0:54I'm going to train several types of models including a causal transformer and also

0:58a GRU The reason why I'm training these is because I think virality has

1:03some sort of sequence dependency

1:05If you're creating many videos in a series then that can affect the videos

1:09that came before it This is unlike something like an image model that you

1:14use to generate images It doesn't really like change over time but I think

1:19some sort of recurrent network makes the most sense So I basically spent over

1:2450 just telling Codex to take all of my data

1:30take all of my TikToks all my YouTube videos all my X posts all

1:34my Instagram posts all my articles everything Take all that data and then it's

1:39going to perform feature extraction

1:42Once feature extraction is done I'm going to train my own model to predict

1:47if I give

1:49it a transcript

1:50is this likely to go viral And I'm going to test it and compare

1:54it to the state of the art models like Opus 5 5 and dbt

1:576 1 Solve And my hope is that my custom model does a better

2:02job at predicting virality than these models and I can use it directly within

2:08my platform sort of like a critic Like if you think of a reinforcement

2:12learning agent it's like the critic in the actor critic framework I can give

2:16it the video see if it's going to go viral and then iterate until

2:19it says yes I eventually hope to even make a generative model I can

2:23generate scripts that can make me go viral generate articles that make me go

2:28viral And I'm going to be using the AI workflow that I've created for

2:31creating videos and use that workflow to just push out a whole bunch of

2:37viral content

2:38It's also going to perform

2:39offline data analysis to see what are the hooks that work what are the

2:43things in the videos that I say that tend to make the video go

2:47more viral So when I initially

2:49that way I don't have to continuously train the model I can train the

2:52model once then use some sort of other offline analysis for virality I'm extremely

2:56excited about this project So excited that I didn't even do my traditional editing

3:00I just wanted to put this video out on TikTok

3:02Let me know if you have any questions down in the comments