Rob Horning: "LLMs don’t produce 'truth,' but they allow individual subjects to navigate through ideology as it manifests in routinized clusters of language, prevalent stock descriptions, statistically grounded associations of concepts, and other empirical artifacts uncovered in the mammoth data sets used to train the models. It may not prevent companies and governments from using this technology to subdue populations and make them more ignorant and exploitable, but it may expose the patterns of language that have historically secured various forms of domination and reconciled people to suffering and injustice"
related: the idea that LLMs are “supracognitive maps”
Horning continues: this "is a main part of what Leif Weatherby argues for in his recently published Language Machines. He invites us to see generative models as capturing the 'best-traveled pathways of language,' the 'tepid mush' that he argues is ideology in action. 'For the first time, we are able to surface ideology quantitatively, to scan the ideological surround,' he writes. 'This provides us with a supracognitive mapping' — mapping that is not oriented by and limited to a specific individualized perspective — 'launching us out of the postmodern era, but not on our own terms.'"
"So in this brief moment, before they fully become hegemonic ideological agents in their own right, assimilated to daily life and at work disguising or normalizing the various operations of ideology that shape our everyday assumptions, how we think and perceive the world, LLMs can show us the biases and associations that are passed off as common sense but in an estranged way, without a specific subject who can be made an alibi for them. We can more clearly see, Weatherby suggests, the ideology at work in our own use of language when it is produced mechanically to meet the sorts of purposes we normally use our own internally produced language for. We can see the mesh of readymade concepts and points of view that make for our 'imaginary relation to real conditions' without simply being always already enmeshed in them, as a condition of being able to have thoughts, of being conditioned to think to ourselves in language. 'We cannot flex and bust out of ideology, which is not a negative condition we might understand but a semiotic surround in which we live, move, and have our being,' Weatherby argues. But LLMs offer a way to explore that surround from a relatively detached standpoint."