Hacker News Reader

Have been working on a list of AI companies that train on user data: https://github.com/skiff-org/skiff-org.github.io/blob/main/b.... Will update the Zoom section but still suspect.

View on HN

What differentiates "AI" from "ML" or other predictive modeling that finds its way into an application? I fit a regression on some customer data last week, am I training an AI on customer data? Is it a matter of intent? Of being public-facing? Of being specifically a generative model?

For me, the main issue is the potential of POI/confidential information being leaked.

I think it's fine using "usage data", but the contents of a private conversation should be considered to be... private.

LLMs and generative image models have shown the ability to leak/reproduce training data. That's a big deal.

The original thinking was generative. That is a good distinction to add. I do think regression/analysis doesn't have many of the same issues, ex. no one would have an issue if Zoom said "we're going to analyze how many minutes users spend on Zoom by time of day"

They obviously do statistical analysis of user data still which I’d argue is in the “AI on user data” bucket. So no I don’t think that applies

Reply on news.ycombinator.com