Let’s break AI

[Admiral Beez]

The AI behind ChatGPT is not very intelligent. I tried to break to below. My last question below was autocorrected by traditional google.

 

[christiesplits]

"Does spandrel belong on a condo?"

 

[afransen]

People will laugh at the foibles of current AI techniques until it surpasses human capability.

This is example is a test that is being widely used to break/evaluate large language models on numeracy. There are many previous tests that we used to laugh at how badly these models performed that they now ace.

 

[Towered]

We laugh now, but...