alt Hacker NewsYeah, but the whole reason clean room RE uses two different teams of people is so you can demonstrate that the second team is not simply reproducing that the original expression -- it would have been impossible for them to do so because they had never seen it.
You lose that if you use an LLM which was trained on the original expression.
To keep the clean room clean, you'd need to use an LLM which you could prove had never seen the original.
Replies
Forgeties79 • today at 7:31 PM
> You lose that if you use an LLM which was trained on the original expression.
>To keep the clean room clean, you'd need to use an LLM which you could prove had never seen the original.
In this true though? It seems the jury is out on whether or not LLM outputs are considered “new” or not legally speaking.
➕ show 1 reply
Yeah, that's why I only argued above that you could use an AI for the first team. I thought the comment you were replying to meant all clean-room is fishy.