natdude

Ohhh, cheers - replacing it now

Wow nice! That's literally the best result I've seen yet!

Fair - the video could be expanded to be way bigger (1000s of posts)

I found 1-2 minutes was decent (30+ sample size on average), and a tradeoff for convenience. Background sampling would be ideal but privacy concerns and technical limitations (especially on mobile devices)

haha IKR!

Hi - I made this baby:

Each post is scored based on a bunch of metrics - yes, is gemini setting attribute values for each post.

These attributes are then all summed and some stats used to create your 'scores' - these are just based on percentile, e.g. top 90% violence = a high violence score.

The health is a weighted sum of the other metrics.

Fair point about 'subjective nature of things'. I've found combining the more subjective scoring with statistics and comparison has made the scoring more grounded

Thanks for the feedback!

Oh cool - its like anti-virus for your brain. AI firewall! haha

Oh interesting - thanks for the upload. Ahh - its the N/A political diversity being seen as 'low' when a null should just have no impact on score.

Thanks! Will work on a fix

Oh interesting - thanks for the upload

Thanks, that's a good point - do you think some analysis of advertising would interest you?

There may only be a couple of ads per 'sample' though, so maybe not enough to be interesting.

UPDATE:: The uploading function was not working for a large number of users - this has now been fixed! (it was an issue with the callback URL not being set in the .env which meant file upload progress wasn't returned to the client).

If you have attempted to use this previously without success, please retry - it should work fine now :)

Yeah I tried for a long time on this! Initially I attempted APIs - but they're locked down and you can't get access to user's feeds.

Secondly, I tried a chrome extension, but these don't work on mobile devices (80% of social media users) so it was redundant (plus very brittle as it depended on specific DOM elements). Also, a lot of content is video/audio, which isn't well captured by static analysis.

A video is a bit annoying to take, but yes, it was the only solution I found!

Newly released! Feedback would be great!