Yacine says AI is still early: too many parameters, too slow, too power-hungry
yacineMTB · x · 2026-07-23
Yacine argues that current AI is still extremely early: models are over-parameterized, learning is too slow, machines are too power-hungry, and our understanding of intelligence is too shallow.
He frames the long-term trajectory as much larger than today’s systems and says that realization is what makes the current stage feel alarming.
More from AGI Musings
- Would top math departments hire a student who solved three hard problems with ChatGPT? — damekdavis · 2026-07-23
- AI education has to move beyond prompts and into context, verification, and judgment — Astrokanu · 2026-07-23
- Gary Marcus says a serious AI backlash could still force a global pause — GaryMarcus · 2026-07-23
- One strong mathematician with an AI agent beats bigger teams of non-mathematicians — AlexKontorovich · 2026-07-23
- Git may become obsolete if AI agents can coordinate through a new communication layer — triviatroy · 2026-07-23
- Reply says scheming may emerge as tasks get more complex — davidmanheim · 2026-07-23