Introducing: The roles of computing
← Back to root(I'm a computer person)
Large language models continue to improve by the day, and their aptitude for generative tasks is hard to refute. The environment continues to suffer, as do, from my perspective, humanity, art, and craft.
The owning class and its sycophants counterprotest by claiming inevitability. “LLMs solve a broad class of problems, and solutions to problems will be built when possible. This is simply the first suitable time and technological context.”
This is a faulty argument and warrants analysis. I will not deny that certain problems invite machine learning as their natural solution; in these cases, I'd readily advocate for the use of a small model of well-defined scope, responsibly trained and placed under the control of a community of stakeholders. But much modern, everyday usage of LLMs falls outside such suitable domains. Ordinary people now reach for LLMs to solve problems that, technologically, were solved half a century ago. Then, why are old, light solutions not readily at hand?
Stay tuned for The roles of computing, a series of posts. Each will identify a problem that LLMs purport to, or do indeed, solve; we'll look at how that problem was solved in the past, technologically but not culturally, logistically, or otherwise. We'll imagine how dormant or niche technologies might democratize parsimonious, composable solutions and therefore participate in a future that steps lightly upon the world and invigorates human agency.
New posts will appear on the Atom feed (huh?)