←back to thread

524 points noperator | 2 comments | | HN request time: 0.016s | source
Show context
saeedesmaili ◴[] No.44490550[source]
After reading this I realized I also have an archive of my pocket account (4200 items), so tried the same prompt with o3, gemini 2.5 pro, and opus 4:

- chatgpt UI didn't allow me to submit the input, saying it's too large. Although it was around 80k tokens, less than o3's 200k context size.

- gemini 2.5 pro: worked fine for personality and interest related parts of the profile, but it failed the age range, job role, location, parental status with incorrect perdictions.

- opus 4: nailed it and did a more impressive job, accurately predicted my base city (amsterdam), age range, relationship status, but didn't include anything about if I'm a parent or not.

Both gemini and opus failed in predicting my role, probably understandably. Although I'm a data scientist, I read a lot about software engineering practices because I like writing software and since I don't have the opportunity at work to do this kind of work, I code for personal projects, so I need to learn a lot about system design, etc. Both models thought I'm a software engineer.

Overall it was a nice experiment. Something I noticed is both models mentioned photography as my main hobby, but if they had access to my youtube watch history, they'd confidently say it's tennis. For topics and interests that we usually watch videos rather than reading articles about, would be interesting to combine the youtube watch history with this pocket archive data (although it would be challenging to get that data).

replies(9): >>44490818 #>>44490825 #>>44491013 #>>44491019 #>>44492764 #>>44493027 #>>44495207 #>>44499820 #>>44501925 #
1. larve ◴[] No.44493027[source]
re o3: you can zip the file, upload it, and it will use python and grep and the shell to inspect it. I have yet to try using it with a sqlite db, but that's how i do things locally with agents.
replies(1): >>44493330 #
2. saeedesmaili ◴[] No.44493330[source]
Author mentions that by doing that they didn't get a high quality response. Adding the texts into model's context make all the information available for it to use.