His skepticism is rooted in logic: At their core, LLMs are state-of-the-art statistical models of how humans talk. If yes, the output from the model should be almost by definition Indistinguishable from human language under any statistical test.
I don’t think this is always argued in good faith; At least some of the debate is initiated by those who wish to maintain inadmissibility for their covert use of the technology. But if you sincerely hold this belief, I present you the following collage:
Image shows approximately 150 Amazon book covers that appear when searching “100000 whys” on the site (Add). Some of these books are bestsellers in the category of children’s literature. You can view a zoomable, full-resolution version Here.
There’s nothing inhumane About any of these titles or covers. At the same time, I probably don’t need to explain to you that you’re looking at the purest form of the AI slop that now fills many non-fiction book categories on Amazon. More specifically, what we’re seeing here is an artefact of the tool being semi-deterministic: if a hundred “authors” give an identical signal to their favorite AI tool – say, “Prepare a reference book for children” – The model will probably give functionally identical output in 80% of the cases.
The similarities in the collage go far beyond the choice of titles: for example, all of the covers of the top line feature a roaring dinosaur in the upper left corner of the design. There are many other clusters in the data as well. Look for the recurring red and white cartoon rocket, a golden retriever, a lion, etc.
This is what makes LLM writing unique: it’s not that the models’ personal mannerisms are different from ours. The point is that they resort to the same, complex methods in response to almost any common signal. This is a vague signal, so you shouldn’t fire your intern for asking them to “It’s not this – it’s that”. But in a more informal setting, it’s okay to trust your gut. In fact, these trends are becoming increasingly important as traditional models of online interaction collapse if producing content takes much less effort than engaging with it.
P.S. If you’re using LLM to automate blogging: Yes, the technology is amazing, but chances are, your publication could be renamed “100,000 Whys.”
<a href