Discussion about this post

User's avatar
Dev Null's avatar

"To avail oneself of the power to outsource articulation to a language machine makes a certain sense under particular conditions, that is to say the conditions that constitute efficiency, optimization, and productivity as the highest human goods."

I might be careful of this formulation since it cedes the argument that LLMs are in fact maximizing efficiency. In my experience they are marginal for pure research due to unreliability unless very tightly constrained and used by someone who already posseses the expertise to catch the inevitable errors. In that role they seem to outperform web search due to web search being so highly corrupted.

I have found LLMs most useful for generating masses of text that I don't much care about. The sort of work related verbiage that no one will read with any engagement and will most likely be skimmed by some bean counter to check a box somewhere. Which is to say the LLM is "efficient" in so far as it allows me to pay this tax of mediocrity and process which in a truly efficient system wouldn't exist. No value is added anywhere in this interaction.

In this sense I would say that the LLMs are currently mostly bandaids that allow us to paper over certain problems in the short term at a fairly high long term cost. Is there any doubt that they will be driving the bulk of call center interaction within a few years? Would this be sustainable without customers being locked in to a small number of monopoly providers that can provide subpar customer service since there is nowhere else to go?

As to increasing real efficiency, there are domains where progress has been made by "AI" such as computer vision but I have my doubts about the long term prospects of these LLMs providing tangible benefits beyond some narrow niches even in the currency of true efficiency and not simple strip mining of value.

Phoenix's avatar

I keep coming back to the phrase owning our words. The thing about language machines is that they can produce language endlessly, but no one quite has to answer for what they say. The sentences arrive polished, plausible, and strangely uninhabited. But the awkwardness of human speech, the pauses, the wrong metaphors, the slow search for the right word, isn’t just inefficiency. It’s where thinking actually happens. It’s where we discover what we mean and whether we’re willing to stand behind it. A world full of fluent language that nobody truly owns feels less like progress and more like a quiet evacuation of responsibility.

24 more comments...

No posts

Ready for more?