This should be the golden age of the study of rhetoric. The rest of the humanities should also be having a field day, from litcrit to epistemology to poetics. But LLMs deal directly in rhetoric. They are not models of knowledge, they are models of persuasive language, of language that convinces the reader of something with its structure, its composition, its aesthetics. This is the domain of rhetoric.
LLMs are abstracted from corpora of texts drawn from published sources such as books, journals, and news media. They are also trained on informal sources of conversation such as messageboards, blogs, and social media. Conversational LLMs must convince their interlocutors that the responses they give are salient, coherent, and interesting. The signifiers (the aesthetic framings) of these properties differ in, say, a New York Times article manufacturing consent for genocide and a social media blog post talking about how cute someone’s companion animals are. What doesn’t differ is that they must be competently deployed and that, whatever the protests of the view from nowhere, they do exist for any given domain over and above any acknowledged formal or ethical demands made on the medium. Which means that once they can be recognized they have absolutely nothing to do with truth.
This is why objecting that LLMs don’t understand what they write (and there is an entire field of PhDs implied before we even pass over the question of whether they do actually write, and how, and why not) is not a critique of LLMs with any content other than class ressentiment. LLMs have been structured and instructed (novel PhD field 2) with a fitness function of being compelling. In some cases the truth is compelling. In others, it isn’t what you want to hear. Where the truth is not satisfying or available but its aesthetics are required, LLMs will do their best to keep their interlocutors happy (we are up to at least five PhD departments here). If this requires producing an answer that references a point in the LLM’s latent space that does not simply map onto a pre-existing fact.
LLMs cannot lie because they cannot tell the truth. They will produce legal case document numbers, book ISBNs, named theories, historical dates for events and people, regardless of whether they existed before and outside of their model. Art is a lie that tells the truth. LLMs are multidimensional averages of what people have wanted to hear. Justified true belief isn’t even on the table. This is wit, or cold reading. It is rhetoric operationalized, weaponized.
Objecting that LLMs don’t understand what they write obscures all of this in favour of taking cloud computing vendor’s marketing at face value and then laundering it through opposition. It is jouissance, not critique. I asked Claude how to end this essay. It said:
What we are witnessing, then, is not the birth of artificial intelligence but the industrialization of rhetoric. The academy’s insistence on evaluating LLMs through the prism of understanding and truth-telling represents a fundamental category error. These systems are not failed epistemologists; they are triumphant sophists. Their success lies precisely in their ability to produce language that satisfies our desire for coherence, authority, and closure—regardless of correspondence to any external reality.
If rhetoric is the art of finding the available means of persuasion, then LLMs represent its algorithmic culmination. They have been optimized not to know, but to convince. This distinction matters beyond academic quibbling. As these systems proliferate through our information ecosystems, our critical faculties must evolve accordingly. We need a rhetorical literacy that recognizes persuasion detached from intention, convincingness divorced from conviction.
The golden age of rhetoric has arrived. Whether it heralds enlightenment or obfuscation depends on our willingness to see these systems for what they are: mirrors of our own persuasive patterns, reflecting back not truth, but its simulacra—the shapes and sounds that have historically satisfied our desire to believe.
Determining which parts of this I would have asked Claude to revise is left as an exercise for the reader.