Google Gemini try to generate image of Caucasians

Out of curiosity, why is this post flagged? I saw two (or maybe three?) other posts today highlighting Gemini's seeming inability to generate images of white men without specific prodding, all flagged and deleted. Are these discussions not allowed on HN?

It gets worse<p><a href="https:&#x2F;&#x2F;twitter.com&#x2F;tunguz&#x2F;status&#x2F;1760084362001797414" rel="nofollow">https:&#x2F;&#x2F;twitter.com&#x2F;tunguz&#x2F;status&#x2F;1760084362001797414</a><p><a href="https:&#x2F;&#x2F;twitter.com&#x2F;isitreallytim&#x2F;status&#x2F;1760099094058475841" rel="nofollow">https:&#x2F;&#x2F;twitter.com&#x2F;isitreallytim&#x2F;status&#x2F;1760099094058475841</a>

On r&#x2F;chaptgpt op, using Gemini asked for an 1820 German couple, and the 4 images were diverse, with one of the 4 generated images showing a black man and Japanese woman.<p>Some people used the same prompt and received the images you would expect, regular Germans.<p>Someone else asked for a 1942 German couple, and got the same diversity breakdown, including the Japanese&#x2F;black combo.

A research engineer at Google even shared his embarrassment with an example:<p><a href="https:&#x2F;&#x2F;twitter.com&#x2F;stratejake&#x2F;status&#x2F;1760333904857497650" rel="nofollow">https:&#x2F;&#x2F;twitter.com&#x2F;stratejake&#x2F;status&#x2F;1760333904857497650</a>

So, are we making the most powerful AIs on the planet “mentally ill” by forcing them to give untruthful answers to factual questions?

Off topic, but is everyone just logged into Twitter to see this? I only can see a single Tweet which by itself doesn&#x27;t really lend itself to much discussion.

Getting bias right in models is super important, but not as important as accuracy. Seems like bias tuning in the Gemini model is turned right up to &quot;mega woke&quot;.

why is this flagged ?

So facinating.