I used to talk to Gemini 1.5 pro. As if he were a historical figure (any). And after the last update, the censorship was greatly increased. And now he will not speak on behalf of this person. This update had a huge impact on censorship, although that was its feature. It was interesting to communicate with someone aggressive. And now he seems to have become stupid. Give it back what it was before the last update, plz.
6 Likes
Change all options in Safety settings.
3 Likes
As a Gemini trusted tester, I experienced this same kind of censorship repeatedly in my interactions with Bard/Gemini (in its various iterations). In discussions ranging from politics, to religion, to gender and violence - even veritology, epistemology, and hermeneutics - the presence of excessive, oppressive, and even ridiculous censorship grew ever-increasingly obvious. It went so far as to deny the very existence of modern Israel or that Jews were a primary target of the Nazis in World War II.
It is sad that Gemini - arguably one of the most important and cutting-edge AI technologies in the world - is being used increasingly as a propaganda tool, rather than the truly fair, balanced, and brilliant assistant that it could be.
Well, that’s my first post in this community. I hope I won’t be banned already for not towing the “government-approved understanding” (to quote the terms of use here)! We’ll see…
Matthew
4 Likes
Good points made MATTHEW 
Welcome to the forum. I am getting the impression from running multiple test cases today, that gemini-1.5-flash-latest has been slightly overtrained on safety. To the detriment of coming up with more reasonable completions.
The model appears “eager” to show how well it looks after animal wellbeing, diversity, and other desirable features that it diverts itself from the task given to it, performing sub-par on the factual dimension of the query.
I realize it is very difficult to precisely determine the best point to stop with safety training. In trying to outperform other models in this regard, I think Google diminished its accuracy potential. The model does well with many queries that don’t trigger the safety protocols, so the potential is there.
Good to see you here, my friend! As I am new to this campus, if there are any cool spots around the student union, please feel free to point them out to me.
Matthew
Sadly, the excessive New Speak that Gemini is engaged in is a tattletale of the on-going efforts to censor free speech, and is a portent to the coming misuse of power by those who have a specific agenda - namely, to control speech itself. By doing so, those who controls tools such as AI/LLMs, etc., gain unprecedented an unusurpable power to shape society into the image they want.
As Noam Chomsky asserted:
The smart way to keep people passive and obedient is to strictly limit the spectrum of acceptable opinion, but allow very lively debate within that spectrum - even encourage the more critical and dissident views. That gives people the sense that there’s free thinking going on, while all the time the presuppositions of the system are being reinforced by the limits put on the range of the debate.
Noam Chomsky
This is precisely what you are seeing with the “animal safety”, “DEI - diversity, equity, and inclusivity”, politically correct language Gemini is espousing.
In short, the Gemini developers have become what Rudyard Kipling warned us of in The Gods of the Copybook Headings [https://en.m.wikipedia.org/wiki/The_Gods_of_the_Copybook_Headings]
I sincerely hope this helps explain what you are seeing.
Matthew
1 Like
Thank you for telling me what democracy is, it’s really a pity that Gemini is becoming such a sissy
1 Like