AI and agents are exploding right now. The most “leftover” area, but with the loudest noise, is all kinds of AI self-styled amateurs—people with no real grounding. This includes, but isn’t limited to, pure vibe coders who invent all sorts of amateur concepts (like loop graphs), disregarding software engineering quality entirely. Representative figures include Peter.

Across major university labs and PhD students, representative figures include the various labs at UC Berkeley, its PhD students, and even old professors. Their papers are full of fabricated stuff: claiming there are a hundred test datasets, but in reality it’s the AI itself scoring and testing itself—no one understands it, and reliability is almost zero. Moreover, the leaders are often completely clueless about software engineering, treating shit like treasure.

A problem solved by a startup in one go can, in this kind of AI academic circle, generate ten thousand “shocking” papers.