Another reason LLMs are not the future of dialog

Executives have been trying to get rid of writers ever since moving pictures were a thing. Samuel Goldwyn allegedly threw a writer out of his office, and told his assistant, "I never want to see him again! Until we need him."

Many years ago, a Canadian network, probably the CBC, decided it could save a lot of money if it made a series about a bunch of commuters taking the train home. The actors would improvise their dialogue based on the day's headlines and whatever else came into their heads.

The show was a ratings disaster, but it was cheap to produce, so it stayed on the air longer than it should have.

The latest idea for getting rid of writers is large language models, aka generative AI. So far, the demos have not been promising. AI-scripted characters turn out to be boring and creepy. 

Well, think about it. If a roomful of live actors can't keep your attention, how can autocomplete software do it, when it has no concept of what it's saying?

Think about improv. The audience for live improv is tiny. If The Groundlings weren't a farm team for Saturday Night Live, who knows if it would exist as a business? Why? Most improv is a stunt. It is like the dog who walks on his hind legs:  it does not walk very well, but it is fascinating to watch it succeed at all. 

It is hilarious to watch the actors interpolate truth based on "yes, and." But you can only watch so much of it. No one is binge-watching improv. 

I don't doubt that there are sublime moments in improv here and there, but you don't see much live improv on TV. Saturday Night Live runs on cue cards. Paul Feig lets his actors improv, but he does dozens of takes and edits a performance together in post.

So if people are not, by and large, watching live human beings making stuff up on the spot, why would we watch autocomplete software do it? 

It remains to be seen how long it takes Hasbro to realize that no one wants D&D scenarios based on autocomplete, either, but I figure we've got another year of this before LLMs implode like NFTs and Beanie Babies.