4 ms·
The fact that the makers of the technology are not able to keep it under control. Copyright infringement is one of the bigger long term risks, and it seems ope
by hyperman1 3y ago
The fact that the makers of the technology are not able to keep it under control.
Copyright infringement is one of the bigger long term risks, and it seems openai aborts the chatgpt output if it returns too much of a copyrighted work.
In this case, chatgpt managed to notice the problem and find a workaround for that filter.
This means chatgpt can escape the security mechanisms that keep it under control. No biggy in this case, but a huge problem if it keeps getting better.
- ChatGTP 3y agoSo we're talking about "alignment" problems and copyright? It seems to be spitting out "Fear Litany" verbatim, so it seems like the model does indeed include copyrighted works. Giving rise to the idea that models actually are a form of compression and do contain copies of copyrighted works.