@mirabilos
In a scientific context, "AI" was prompted to "find highly toxic chemicals" (not exact wording).
It found some 40.000+ combinations, many of them also highly explosive, if I remember correctly.
Scientists themselves would have never thougt about those.
After some debate, whether to publish them, they were published.
https://www.theverge.com/2022/3/17/22983197/ai-new-possible-chemical-weapons-generative-models-vx
---
How far "in the centre of the box" is this example, in your opinion?
@fasnix @pluralistic that is of course relative to the total size of the box and the context; I’m sure they would not use somethint like ChatGPT for that but something specialised.