World Models Won't Get Us to AGI — Continual Learning Is the Missing Piece, and It's Hard
Intelligent-Cream-14 · reddit · 2026-09-16
The author argues world models alone won't reach AGI/ASI; continual learning is the only missing piece — and likely far harder than it sounds.
- The update mechanism must be more complex than the model itself: like the brain, it must decide when to update, by how much, and what to forget/merge to free capacity.
- Today's models win via RL on verifiable outcomes, but continual learning is hard to verify — essentially 'RL on an RL algorithm,' possibly stacked many layers deep in human brains, where even reward definitions get updated.
- Practical pain point: three months of tacit experience can't be implicitly inherited by a model; you must write it into prompts, while an average colleague picks it up naturally.
- Verdict: good news that it's the last piece, bad news that solving it may be brutally hard — or force an entirely different path to AGI.
More from AGI Musings
- Databases to distributed systems: every systems skill is merging into AI inference and hardware — Abhishekcur · 2026-11-02
- Limitless analogy: smarter people may gain far more from superintelligence — McDonaghMatthew · 2026-11-02
- Hunting for the most concrete "curing cancer with AI" roadmap of 2025 — iskander · 2026-11-02
- Avoiding AI brain rot: Miles Cranmer's lecture on learning with AI — _ScottCondron · 2026-11-02
- Yacine mocks "autonomous research" hype as Naam argues RSI loop needs 5-10x boost to run away — yacineMTB · 2026-11-02
- Vitalik's "post-AI cryptography": pseudonyms are dead, per-action anonymity is in — devanshmehta · 2026-11-02