So the Rosenbaum FUTURE OF TRUTH thing is appalling, but let’s not pretend this is an aberration: the over the fence talk I hear is that full on AI is now the norm in trade nonfiction.
A thing we’re seeing with the encroachment of AI on services like Google search is a highly counterintuitive displacement of the list—an information genre honed over millennia—by discursive prose. In the distinction popularized long ago by Lev Manovich, narrative is triumphing over database.
At an online event this morning I defended a claim which others found troublesome: That there are legitimately experienced aspects of computational phenomena, such as large language models, which—for all practical ways of speaking—are not always resolvable by or restorable to material conditions.
I spent the first half of the day at a public facing AI convening in downtown DC. Eclectic speakers, collected audience. The biggest blindspot I saw was this: Culture was repeatedly positioned as something *outside* of tech. We can make better tech if we bring more culture into the process. +
I understand why we’re still explaining to people that LLMs are not conscious but I also don’t understand why we’re still explaining to people that LLMs are not conscious. Like imagine thinking this is the conversation to have.
Does anyone know of an account of what was happening inside of Google‘s DeepMind c. 2015-2020? Looking for something like Karen Hao’s work— Definitely doesn’t have to be book length, but that combination of investigative reporting and narrative history.
Despite the frequency of the analogy, one should keep in mind that autocomplete and LLMs are materially different technologies. Just think about how willfully stupid the services built into your phone can seem (for example). Case in point: 🧵
Meanwhile back in my mentions, for the second day running, a white, male, and doubtless very progressive minded lad who has read himself a great lot about “bullshit machines” on this website continues to explain to a credentialed (female) computer scientist how generative AI really works.
In a talk I’ve been giving this spring, I’ve pointed out that “the same authors who are tireless critics of the tendency to anthropomorphize large language models by way of words like ‘know’ and ‘understand’ and ‘think’ seem content to dismiss them based on their inability to ‘mean’ or ‘intend.’” +
Calls for refining our vocabulary around “AI” often seem to originate outside a space of social use. Language allows for precision, but it is also a form of shorthand, a compression algorithm if you will. Which is to say language is metonymic, and thus slippery. I doubt this can be overcome by fiat.
