You included all the proof we need in your comment. Grok has been proven to do it, others may be doing it as well. And if its isn’t promoting fascism maybe its promoting some other ideology. The point is these models are not unbiased, and in many cases are being manipulated by their owners.
It’s not maybe they have been researched and are highly manipulable.
You are trusting the host of the model to not introduce biases. Getting the model to regurgitate it’s hard coded prompt info has already happened and different providers have been doing that
You included all the proof we need in your comment. Grok has been proven to do it, others may be doing it as well. And if its isn’t promoting fascism maybe its promoting some other ideology. The point is these models are not unbiased, and in many cases are being manipulated by their owners.
It’s not maybe they have been researched and are highly manipulable.
You are trusting the host of the model to not introduce biases. Getting the model to regurgitate it’s hard coded prompt info has already happened and different providers have been doing that