or depicting right-handed individuals when asked to show someone writing with their left hand. These tendencies arise from the abundance of such imagery in online data and can only be overcome through extensive reinforcement training.

Go ahead and try it out!

Thanks Mark for sharing this!
Control over the "trusted sources" of AI models will determine ultimate authority.
And AI models are susceptible to manipulation through such sources.
For example, AI image generators frequently default to showing watches at 10:10

or depicting right-handed individuals when asked to show someone writing with their left hand. These tendencies arise from the abundance of such imagery in online data and can only be overcome through extensive reinforcement training.
2/2