Are continual learning, causal reasoning, world models, embodied experience, persistent memory, or social understanding still major unsolved problems?
LeCun and others argue autoregressive LLMs structurally lack world models and continual learning; scaling proponents reply that these emerge with scale or can be scaffolded around. Continual learning is increasingly cited as the binding gap.
View on the map → · Open in Browse →
The missing-ingredients list got both shorter and better funded: a companion analysis to the Definition-of-AGI paper argues long-term memory is the only gap where models score zero, Google shipped an architectural attack on exactly that gap, and LeCun raised over a billion dollars on the thesis that the whole list — world models first — requires leaving the current paradigm.
the rest of the way will mostly require business-as-usual research and engineering
Scores GPT-5 at ~57% of AGI and argues every remaining gap except one is tractable engineering — long-term memory storage/continual learning is the only capability where models 'score zero' and a genuine breakthrough may be required.
LLMs are limited to the discrete world of text. They can't truly reason or plan, because they lack a model of the world.
Well-reported account of LeCun's AMI Labs and his JEPA/world-model program — the most prominent institutional bet that the missing ingredients require a fundamentally different architecture.
Nested Learning treats a single ML model not as one continuous process, but as a system of interconnected, multi-level learning problems that are optimized simultaneously.
Google's flagship architectural attack on catastrophic forgetting — the main evidence that labs are treating continual learning as solvable by architecture rather than scaffolding.
The models today are pretty capable, but there are still some missing attributes: things like reasoning, hierarchical planning, long-term memory.
Lab leadership's current missing-ingredients list — continual learning, world models, hierarchical planning, long-term memory, and hypothesis invention — notably longer than scaling-optimist rivals concede.