Building an AI to Predict My Next Post Performance

0:00Here's how I plan to create an AI that will tell me whether a

0:03TikTok that I'm about to post will do any good before I even post

0:09it Step 1 I'm going to take all of my media content from all

0:13my social medias So this includes TikTok but it also includes any videos I

0:18posted on X any videos I posted on Instagram Reels and any videos I

0:22posted on YouTube Step 2 I'm going to use other inputs as well So

0:26if I posted carousels on TikTok or Instagram I'm including that as well And

0:30I also have hundreds of different articles that I've written on my Medium and

0:36on my personal blog Then I'm going to send the media to a language

0:39model that understands the media type So for example for videos I'll send it

0:43to a model like Newspark 1 2 or Gemini Flash 3 8 because these

0:47models understand how to process videos And the output of these models will be

0:52a JSON list of features And it might have things like is the hook

0:56a question Is the hook a statement

0:59Or it might have things like b roll is present in this video Or

1:02screenshots are present in this video And there's a list of features from the

1:07video Then I'm also going to send the inputs into an embeddings model as

1:10well

1:10Embeddings model like Gemini Embeddings 2 which understands multimedia inputs I'm going to use

1:16this for all media types for all my different videos and all my articles

1:19and photos And then I'm going to train a causal transformer that can predict

1:25how many views or likes the videos get after one week The reason why

1:32I'm doing this fixed time frame is so that the comparison is roughly equivalent

1:36So for example if the videos get a thousand likes after a month versus

1:41a thousand likes in a week those are two different outcomes right So I

1:44want a fixed time frame And then to train a transformer I'm also going

1:49to train a GRU

1:51Because constant virality is sort of sequential The series of videos you post might

1:56have an impact on the virality of the next video So I want to

2:00train some sort of recurrent neural network Then I'm going to train all these

2:04networks And I'm going to train an ensemble network I'm going to see if

2:08these networks can predict virality better than using a model like Opus 5 5

2:13or GPT 6 1 After doing all this if I have interesting results maybe

2:18I'll write a paper or something I don't know We're just going to have

2:21to see