Reddit debate: neural network fundamentals never changed — is 'we don't understand AI' hype?
nezvanovova · reddit · 2026-09-25
A Reddit user with CNN research background challenges the popular claim that "we don't understand how AI works": gradient descent, parameter generation in multidimensional space and convergence haven't fundamentally changed since handwritten digit classifiers; papers describe architecture and attention changes, not new core concepts. The thread sparked debate over whether 'understanding' means mathematical describability or human-intuitive interpretability of emergent LLM behavior.
More from AGI Musings
- Jeff Sebo backs Nate Silver: most AI experts would agree models 'think' — aran_nayebi · 2026-09-25
- DeepMind CEO stands by it: today's systems are nowhere near AGI — luislamb · 2026-09-25
- The most plausible AI doom scenario: AI in control of everything, humans nothing — kevinnbass · 2026-09-25
- A 1982 book predicted 'application development without programmers' — and it reads like today's AI debate — ZeeshanZiaML · 2026-09-25
- A battery PhD used AI to map a decade of research — do we still need to read 100 papers? — Clean-Interview-5081 · 2026-09-25
- Neurosymbolic researchers argue LLMs can't do abduction alone — luislamb · 2026-09-25