Beyond that, Google’s argument was deemed particularly weak, since the AI overview in this case “contains statements that do not appear in the search results at all.
This is going to have some pretty far reaching and large consequences.
Essentially, it’s impossible to remove pretrained “sentiment” from the response at query time, or post training or fine tune time. The model is trained on a bunch of data including Reddit and other social media, and good luck pruning that effectively.
But this also implies a statement made in the Gemini app, or ChatGPT, or any LLM which is not strictly citing RAG data is going to be liable for slander.
If you ask an LLM if Enron was an honest company, it’s going to say no. If Enron were in Germany, they could sue over that.
I’m not saying this ruling is wrong. Companies can’t just “it’s for entertainment purposes only” their way out of any liability, that’s asinine.
This is going to result in LLM companies having to cut providing content in Germany.
This is going to have some pretty far reaching and large consequences.
Essentially, it’s impossible to remove pretrained “sentiment” from the response at query time, or post training or fine tune time. The model is trained on a bunch of data including Reddit and other social media, and good luck pruning that effectively.
But this also implies a statement made in the Gemini app, or ChatGPT, or any LLM which is not strictly citing RAG data is going to be liable for slander.
If you ask an LLM if Enron was an honest company, it’s going to say no. If Enron were in Germany, they could sue over that.
I’m not saying this ruling is wrong. Companies can’t just “it’s for entertainment purposes only” their way out of any liability, that’s asinine.
This is going to result in LLM companies having to cut providing content in Germany.