The Bullshit-Industrial Complex just got Bullshitier
Generative AI poses an acute short-term risk to information quality on the internet.
2022 has been the year that Generative AI has truly come into its own - it has evolved from 'hmm, that's interesting' to 'useful for some narrow use-cases'. Unfortunately, the incentives to adoption align perfectly with those who create low quality content and lots of it.
Prediction: the scaled output from LLMs will soon pour rocket fuel on the furnaces of the Bullshit-Industrial Complex.
Deconstructing, this implies that LLMs will be adopted and adopted quickly. And that on average LLM-augmented writing will be worse (in style, usefulness or truthfulness) than those illustrious few writing outside the Bullshit-Industrial Complex.
What has me confident in this pessimistic outlook are two core assertions:
LLMs do not need widespread adoption to flood the internet. They are fast and cheap. With the means of production automated, a cohort of early adopters can quickly outproduce trad writers.
The most-incentivised users create the lowest quality content. The major tech platforms reward the regular and frequent creation of content. Search rewards (amongst other things) hyper-specific content (with affiliate marketing piggy-backing on that).
There are of course counter-arguments to this prediction.
Perhaps some second order effect will now drive readers to more thoughtful curation of sources, or search and recommendation will be forced to learn better ways of surfacing the most relevant content. Or we may learn that humans are naturally better editors than blank slate writers - and so, article quality may instead improve.
I hope this is true.
At a 3-5 year time horizon, I expect that users and the algorithms that feed us will adapt to the coming tidal wave of machine-generated content. Going further, it may even be a good thing and hasten the demise of the Bullshit-Industrial Complex.
And there are, of course, other use-cases where the next generation of LLMs in particular are expected to excel.
But in 2023 and 2024, I wear my pessimism hat proudly and expect that the average quality of written content online will nose-dive.