Hacker News Reader

There's two decades worth of countless conversations on Reddit alone that would be buried into nothingness but instead ML has revived all that activity as useful data. ML is definitely a great way to bring back utility for a lot of old and unused data.

View on HN

> revived that activity as useful data

Revived as compressed text associations, it is potentially useful data, but also potentially totally wrong in non-obvious ways. (Or, to riff on Futurama, "The worst kind of incorrect.")

It is used to help train the LLMs on how to "talk" like normal people, even if the topic they're discussing isn't that useful or valuable.

This seems like a reasonable take to me. I wish those downvoting you would explain where they disagree.

Reply on news.ycombinator.com