Working Up

by Dwayne Phillips

Let’s stop hyperventilating about this and that AI thing. The marketplace will decide and with more wisdom than myself.

I just read yet another article about yet another artificial super-duper large language model.

How can a model of something be larger than the something? Isn’t a model a smaller representation of something? I digress again to the English language. I must stop doing that.

Back to the subject, this brand new model was “hallucinating.” It was creating images of people with three arms and four knees and two fingers and all sorts of things that didn’t make sense. Like we used to say, “It was full of errors caused by the people who built it.”

We don’t like to say “errors” much anymore. We don’t like to attribute these errors to the people who built something much anymore. Hmm. I guess that says something about us.

Back to the subject, someone built a bad product that is full of errors. Remember the Yugo car? Remember Ken Burns’ documentary on baseball? Just two examples of products full of errors. The marketplace rejected them (eventually).

The marketplace will reject these AI things that are full of errors. Maybe someone will build an AI thing that doesn’t have (as many) errors. We know that anything built by people will have errors in it because, well, us folks are error prone. Sorry about that. If we could just build a machine that builds machines then those machine-built machines won’t have errors in them because they weren’t built by error-prone people. Sorry. People are in the chain of events somewhere.

I suppose there is value in someone writing an article that scoffs at a product that is full of errors. Perhaps that speeds the marketplace’s rejection of the product. Perhaps not as, “there is no bad publicity” or something.

The marketplace will reject the error-filled product. Let’s try to reduce the number of errors. We can do better.