Taking Stock of the Current LLM Hype Cycle
We’re now several years into the LLM hype cycle and the
volume at this point is so noisy and disjointed that it feels
muted, like having water in my ears. As an act of survival I have
complied with what is expected of me: paying my bills is a far
more important act than sticking to my principles. Putting a roof
over my head and keeping myself in line with the world around me
trumps my opinions on anything. I use AI at work, and it’s
unlocked certain types of productivity at the expense of others.
Thought still works at the same pace, so pure volumes of
language-like symbols put out into the world on random walks can
only help in places where we need more words, any words, and not
the right words. Many strategies, from models with thinking stages
to entire safety harnesses seem to reason that the only way to
handle the volumes of words put out by these random walks are
equally voluminous counterstreams of symbols to keep them at bay.
The solution to most LLM problems at most AI-forward shops is,
surprising no one, the indiscriminate application of more LLM.