When A.I.’s Output Is a Threat to A.I. Itself | As A.I.-generated data becomes harder to detect, it’s increasingly likely to be ingested by future A.I., leading to worse results.

silence7@slrpnk.net · 10 months ago

When A.I.’s Output Is a Threat to A.I. Itself | As A.I.-generated data becomes harder to detect, it’s increasingly likely to be ingested by future A.I., leading to worse results.

jet@hackertalks.com · 10 months ago

The best analogy I can think of:

Imagine you speak English, and your dropped off in the middle of the Siberian forest. No internet, old days. Nobody around you knows English. Nobody you can talk to knows English. English for all intents purposes only exists in your head.

How long do you think you could still speak English correctly? 10 years? 20 years? 30 years? Will your children be able to speak English? What about your grandchildren? At what point will your island of English diverge sufficiently from your home English that they’re unintelligible to each other.

I think this is the training problem in a nutshell.