Gary Marcus

<start of post>

The AI "productivity miracle" is hitting a wall of reality.

We’ve been told for eighteen months that LLMs would automate the middle class into a life of leisure. Or, at the very least, make us 40% more efficient by Tuesday.

But the data is starting to tell a different story.

Integration costs are skyrocketing. The "hallucination tax" is real—meaning you spend more time fact-checking the machine than you would have spent writing the draft yourself. And then there is the issue of morale.

Nobody wants to be the "human-in-the-loop" for a mediocre stochastic parrot.

Mark my words: The companies that win the next five years won't be the ones that replaced their staff with GPT-4 wrappers. They will be the ones that used this pause to double down on human expertise.

The soaring cost of compute vs. marginal gains

The legal minefield of training data copyright

The "slop" problem degrading the quality of the open web

It is much easier to be the cheerleader. It is much harder to be the one pointing at the empty tracks while the hype train is at full speed.

Where are my pom poms?

*

Full analysis of the "hallucination tax" in the comments.

#artificialintelligence #technology #innovation #economy

<end of post>