Google's hidden AI diversity prompts lead to outcry over inaccurate images (arstechnica.com) 13 points by LinuxBender 2y ago ↗ HN
[–] tylrprtr 2y ago ↗ There's something humorous to me about an outcry over generated, fictional images being somehow "inaccurate".
[–] mjbeswick 2y ago ↗ Bias exists in training data generally because it reflects reality.Considering the example give the article, the average CEO is a white male; so surely you would expect that unless you explicitly asked for diversity!?It's should be down to users to alter their prompts if they wish to generate images outside of statistical norms. [–] Piko 2y ago ↗ I disagree. We aren't telling kids "With your parents' socio-economic background and skin color you cannot become whatever you want.", after all. [–] name_nick_sex_m 2y ago ↗ ideals vs reality [–] add-sub-mul-div 2y ago ↗ That's an easy way to launder our bias of choice to algorithms, but thankfully we're not falling for it.
[–] Piko 2y ago ↗ I disagree. We aren't telling kids "With your parents' socio-economic background and skin color you cannot become whatever you want.", after all. [–] name_nick_sex_m 2y ago ↗ ideals vs reality
[–] add-sub-mul-div 2y ago ↗ That's an easy way to launder our bias of choice to algorithms, but thankfully we're not falling for it.
[–] Triphibian 2y ago ↗ When a company puts secret prompts into their AI it should be called "Robocopping."
8 comments
[ 4.0 ms ] story [ 28.1 ms ] threadConsidering the example give the article, the average CEO is a white male; so surely you would expect that unless you explicitly asked for diversity!?
It's should be down to users to alter their prompts if they wish to generate images outside of statistical norms.