The cost of growing and running large LLMs far exceeds the economic value of small task automation.

Small, targeted, on-device, cheap to run models make sense in some places. General purpose LLMs that take many gigawatt data centers to lose money producing questionable output don't.

We are finding ourselves simultaneously trying to divest from fossil fuels while increasing our energy consumption faster than ever. What are we getting for it? Slop, sycophantic rubber ducks, and merge conflict Clippy.