I wrote a script to analyze chat exports

[pablo_aka_god]

I've been having manny misunderstandings with my GF and because there is a lot of gas lightning going on and in my pursuit to get more perspectives on the issue I exported our entire 8 month whatsapp conversation and wrote a script that would analyze the whole thing and write a report with the personality type of each other, our pros and cons. This way we could get an outside opinion on our entire dynamic.

Of course LLMs have their own biases, but this is another datapoint. 

I'm sharing the code, for the nerds out there that want to run it just replace the variables at the top with your own api keys and name of the txt export with the whatsapp/other chat app export in your google drive.

Link: https://colab.research.google.com/drive/1hrytnWX71uajPVaRNgMYtjhbDizQ14jw?usp=sharing

 

Why use this code and not just upload the entire convo to chatgpt's web app ? because convo is too large and my code chunks it in parts so that the AI can go through the entire thing.

 

Let me know if you use it!

Edited Wednesday at 03:56 PM by pablo_aka_god

[OBEler]

@pablo_aka_god great idea but outdated. Google Gemini 2.5 can handle 1 million tokens. So you can give it 2 books full of text ab it has still 90% accurate analysis.

Edited Wednesday at 11:22 PM by OBEler

[LordFall]

@pablo_aka_god What conclusions did it come up with?

I think AI counsellors and coaches are the future, we all need one if not many. 

[pablo_aka_god]

@LordFall it was really good for finding personality traits and values, eg: she is more emotional and reactive, he is highly analytical, etc. 
The problem is that LLMs are highly biased, I noticed she was portrait as higher than me only because she would blame me all the time, so the LLM reads that and it fires parts of the network to think badly of the person being blamed,it does not have the reasoning power of a smart human to not trust everything it reads. eg: GF calls me manipulative. then LLM assumes I'm manipulative bc it's just doing a summary of what it reads but actually she's manipulating me by calling me manipulative.

 

I think it's a promising tech but would need way more investment to develop a trustworthy tool. 

Humans are complicated
 

Edited Thursday at 08:08 PM by pablo_aka_god