This is an I-Heart podcast.

Guaranteed Human.

Quarzone Media.

Hi, I'm Ed Zitron and welcome back to Better Offline.

As I've discussed in my last episode, there are four intractable problems that are going to stop Generative AI from going much further.

Its massive energy demands, its massive computation demands, its hallucinations when it authoritatively tells you something that isn't true or makes horrible mistakes in images,

and the fact that these large language models have this insatiable need for more training data.

Yet I think what might pop this bubble is a far simpler problem.

Generative AI simply does not deliver the magical automation that everybody has been fantasizing about.

And I don't think consumers or enterprises are actually impressed.

A year and a half after launch, it seems like the kind of immediate and unquestioning infatuation with chat GBT, and for that matter, other generative AIs, has softened.

Instead, there's this rising undercurrent of apathy and

mistrust and, of course, failure that's kind of hard to ignore. In June 2023, traffic to ChatGPT's

website, where people access the ChatGPT bot in a web browser, fell for the first time since

launch, starting a trend that's

continued for five of the following eight months, according to data from similar web.

People are becoming more aware of the technology's limitations, like as I mentioned, hallucinations,

which, as I know, is when chat GPT confidently asserts things that aren't true, which can be in writing when it gives

you an incorrect fact, or in an image when it gives a dog 18 legs.

To make matters worse, according to data from data.AI, which used to be known as App Annie,

chatGPT's downloads in iOS had begun to drop from a high of just over 700,000 a week

to a plateau of around 450, just over 700,000 a week to a plateau of around

...