As training scales up, labs will only spend compute on math and coding data, researcher argues
moultano · x · 2026-09-18
Responding to the claim that "learning AI" is a path to stable employment, an industry researcher argues:
- Early on, labs used data somewhat at random because it was available, and fields like art "caught strays."
- As scaling ramps up, labs no longer waste compute: data unrelated to building AI—mainly math and coding—is being ignored.
- Implication: the world only needs so many AI builders, and the trajectory points toward recursive self-improvement rather than broad job creation.
More from AGI Musings
- AI is scrambling wages: blue-collar jobs hit $200K-$500K as white-collar work dries up — bindureddy · 2026-09-18
- Gary Marcus: Judea Pearl's causality challenges remain unsolved in the LLM paradigm — GaryMarcus · 2026-09-18
- "Who defines alignment?" AI safety's ideological ambiguity sparks debate — dbasch · 2026-09-18
- Researcher: no good falsifiable scientific test for phenomenal experience may exist — burny_tech · 2026-09-18
- Pedro Domingos: "I'm more worried about Anthropic than AI itself" — acmoytoy · 2026-09-18
- Creatives face 'create and starve' dilemma — and that's why AI is taking their jobs — im_mansigupta · 2026-09-18