GPT-5 and AI Winters

There are better authors than me to talk about this, but historically in AI there have been many boom-overhype-bust cycles. LLMs if they do eventually go “bust“ won’t be the last. The history is ripe with many examples of scaling “learning“ and trying to define tests for intelligence. Every time we haven’t quite gotten there. The current cycle is the first in the popular media.

It seems that we didn’t get recursive self improvement this year like some thought (and hyped, and I wrote about in a way that was hopeful, and was wrong). We might be better another “plateau“ but when it comes to expectations that manifests in a bust. If GPT-5 is a signal socially, the slow march forward continues but the exponential expectations might not be realized. For now.