ML PhD: AI won't destroy the world soon — compute cost is the physical limit
Fit-Gas-5760 · reddit · 2026-09-25
A data-science PhD with 10 years of ML experience argues LLMs are mimics without intent: early 'evil assistant' turns stem from triggered story patterns, not goals. He cites DeepSeek's 2025 paper (arXiv:2503.05132) on surprise-driven reasoning as a strong training pattern. The Transformer paradigm is fundamentally unchanged since 2017, and the real constraint is hardware — he estimates Anthropic's Opus has 5 trillion weights, with compute needs growing faster than inference gains. Conclusion: near-term AI doom is blocked by physics and economics, not alignment.
More from AGI Musings
- Ross Girshick posts CVPR slides recounting 1.5 decades of object detection and its link to the LLM era — CatAstro_Piyush · 2026-11-11
- Ben Todd wrestles with a 50:50 chance only 18 months remain of 'the human era' — ben_j_todd · 2026-11-11
- Mathematicians face identity crisis as AI solves problems in hours, Corriere interview warns — ValerioCapraro · 2026-11-11
- Why inference speed will be AI products' biggest competitive edge in 1–2 years — itsOmSarraf_ · 2026-11-11
- AI Defense Shifts to Scale Economics: Attacks Buried in Legitimate Traffic — DavidLinthicum · 2026-11-11
- Ben Todd's alt log-log plot shows task horizons hitting a singularity in under 12 months — ben_j_todd · 2026-11-11