Imagine sitting there and reading that you killed your 2 kids (sitting in the next room) and tried to kill your third child? That is what happened to Norwegian Arve Hjalmar Holmen.
Now privacy group myob has got in on the fight to try and get OpenAI to clean up its act and follow the law - it is illegal to output incorrect information about real persons.
OpenAI’s response is to shrug and put a disclaimer that the answers could be wrong… The problem is, that is still illegal in the EU, under EU law, if the information may be wrong, it has to be verified and corrected or deleted, outputting incorrect information regardless is illegal.
I feel like something of this level needs to be properly explained. This should not happen under any circumstances, there should be NO room for an AI to make claims someone committed a crime. If there is any reason for it to have acted this way, such as something defamatory in the training inputs, then that needs to be rooted out and exposed and explained.
There have been other cases, allegedly, looking at the comments on the article, an AI reported that a crime reporter who took notes in court and wrote them up for his paper/website actually committed those crimes he reported on.
If you have a government of liars, cheats, and con-men who laugh at the thought of accountability is it a surprise that a corporation would do likewise. Perhaps Europe can come to our rescue.