Plausible Nonsense

My mom forwards me a good financial newsletter every month.

I read something interesting in it this month:

AI models are optimized to deliver plausibility and fluency, but not necessarily truth. In their training, they are rewarded for sounding right, even if not being factually right.

The output is sometimes called “plausible nonsense” – answers stated with great confidence but made up.

That made me think almost immediately of what we hear from defense doctors.

On the surface it sounds compelling. But once you go a little bit deeper, it begins to fray at the edges and unravel. The sun is not a ball of fire pulled through the sky by a chariot.

#plausible

#nonsense

#category