Hacker News Reader

HNHacker News

Top New Best Ask Show Jobs

Comment by transformi | Hacker News Reader

Full thread

transformi·

There were many more already week ago... (location & identity restored from trained data..)

Causing even more (privacy) concerns..

https://twitter.com/MetaAsAService/status/170679883460343414...

View on HN

If that information is easily searched, what's the risk here? I'm not sure I see the harm in a computer being able to identify the high profile owner of a social network or the well-known subject of a popular Internet meme. Guessing locations based on images is literally the premise of the popular game "Geo Guesser".

This is very serious. The fact is that they promoted that they not allowing revealed faces in their paper. That's mean that every person that in the training data could be revealed, and there is not transparency about their training data who is in or not (same as all the artists that are forced to opt-in to their data/ scraped by the GPT-BOT....)

> That's mean that every person that in the training data could be revealed

The examples you linked do not show that this statement is true. Far from it. You showed that one of the most high profile people in the world can be recognized. There are millions (ten of millions? hundreds?) of photos of Elon Musk out there. He was TIME's person of the year. You'd need to show that photos of non-uber-famous individuals who are seen in a small handful of public photos are identifiable for me to believe that statement.

Has anyone made an image bot trained with street-view data? I'm sure it would rely too much on meta-info to be useful outside of geo guesser, but would be swell to see. To that end, has anyone attempted to scrape much of that data google et-al collected?

The link is useless without an account

Try using Nitter as an alternative frontend: https://nitter.net/MetaAsAService/status/1706798834603434149

Reply on news.ycombinator.com