2026-09-06 16:02 UTC
The problem with hoping that we get a "reset" of sorts for software development after the bubble pops is that the shape of LLMs as a tech today is an epiphenomenon of the bubble
Because of the bubble distortions and wildly irrational economics, the post-equilibrium models will inevitably be very different from the current models
This is likely the peak effectiveness of the technology, as right now its backed by the wealth of everything.
Replies (1)
-
@baldur@toot.cafe 2026-09-06 16:03
One core problem is that the tech is volatile. You can get very different models from training run to run even using the same data. Current models can do what they do is because they keep throwing money at runs and training data Because of this volatility, this isn't a tech that works like a foundation. It has lottery dynamics