Discussion about this post

User's avatar
Bo Parker's avatar

If the goal is to “explain,” the fight with AI may be lost as explaining figures to be about what is known and documented broadly - in other words already in the training data.

I recently bought a new car that was a brand new model. As a “rolling computer” it required a complicated configuration process. I asked Gemini for help. It could not help itself - it kept giving me the wrong advice derived from older but similar models of the car. This even after I told Gemini to limit its advice to my specific car.

As a stochastic parrot it was driven by the numerical dominance of older/similar car models in its training data.

So Brad, maybe think first “what’s NOT in the training data?” Yes, it’s ephemeral but also a guide for how to be more than AI.

If it’s not in the training data but is more than wild speculation what is it? The next 5 years of new PhD dissertations in economics?

Thomas L. Hutcheson's avatar

A problem with a paywall, however, is that few people are willing to pay to disagree with the writer, so neither the writer nor the other readers benefit from reasoned disagreement.

16 more comments...

Ready for more?