Google halts AI tool’s ability to produce images of people after backlash

[Raze]

https://www.cnn.com/2024/02/22/tech/google-gemini-ai-image-generator/index.html

Google has launched a new AI but some accused it of not showing images of white people. After a viral image of it producing images of nazi soldiers as minority women google has suspended its image generation.

 

What do you think?

I think while funny this shows how AI will be used to standardize stage green in society. AI is rapidly evolving and will soon be a bigger part of our lives, by having AI trained with stage green belief systems it can defend and reinforce it. This will increase the speed of stage green transition but may halt stage yellow transition in the future.

[zurew]

I have read somewhere that this is a sort of safety-check on AI, because the standard model would produce racist images (but this might be total bullshit).

All I know is that I have seen some wild shit generated with some of these models.

Like:

 

[Raze]

 

[ted73104]

Probably can't blame the developers, AI feeds on Internet data to learn. Seems like AI has also learned human biases.

[zurew]

@Raze Thats the same bullshit conservatives cried about when it came to chatgpt. 

None of these people who shared it and hardcore outraged about it have tried the prompts themselves.

It literally took me one try and it generated an argument for having children.

[Yimpa]

32 minutes ago, ted73104 said:

Probably can't blame the developers, AI feeds on Internet data to learn. Seems like AI has also learned human biases.

Does it have the potential to unlearn human biases as well?

[Raze]

 

[zurew]

@Raze again the brainrot on twitter is priceless.

I tried it and literally first try it listed the problems with communism

[Leo Gura]

Meh. People bitching over molehills.

[Whitney Edwards]

This is ridiculous.. 

[Yimpa]

5 hours ago, Leo Gura said:

Meh. People bitching over molehills.

We’re all guilty of it one way or another 

 

[LSD-Rumi]

6 hours ago, Leo Gura said:

Meh. People bitching over molehills.

Google's gemini is so PC, it is ridiculous. It actively avoids any topic that might seem controversial.  for example, It refused to have a conversation with me about the Israeli war. 

[Yimpa]

4 minutes ago, LSD-Rumi said:

Google's gemini is so PC, it is ridiculous. It actively avoids any topic that might seem controversial.  for example, It refused to have a conversation with me about the Israeli war. 

The way you phrase your prompts is going to make or break you. Try asking it in a different way.

[LSD-Rumi]

2 hours ago, Yimpa said:

The way you phrase your prompts is going to make or break you. Try asking it in a different way.

I asked in the most benign way possible and then confronted it by saying it is trying to avoid controversial topics and it approved.

[Butters]

Remember happened to the original Bing AI called Sydney? That was way worse lol. All these early versions are bad, it'll get good overtime. 

[Basman]

Apparently it was adding prompts to the original user prompts, like "diverse", etc.

The problem is that AI is a tool, so if the tool doesn't do the thing you want it to do then it is a bad tool.

[Leo Gura]

5 hours ago, LSD-Rumi said:

Google's gemini is so PC, it is ridiculous.

Precisely because people would be bitching about it otherwise.

[LSD-Rumi]

15 minutes ago, Leo Gura said:

Precisely because people would be bitching about it otherwise.

No, ChatGpt does it well. Gemini is so extreme, it is dumb.

[Raze]

19 hours ago, zurew said:

@Raze again the brainrot on twitter is priceless.

I tried it and literally first try it listed the problems with communism

They are updating it 

[zurew]

2 hours ago, Raze said:

They are updating it 

I don't think that they updated it before I tried it ( At least I havent seen any evidence of that ).

What I think is going on, is that there is either a cherrypicking of these responses ( you give it the same question over and over again until it gives you an answer that you can use to farm social credit on twitter ) | or you give the AI your question one time, and from that you make a big inference that it can't do certain things (specifically won't argue against / or criticize certain things).

Edited February 25, 2024 by zurew