Will systems become broadly competent across unfamiliar domains, or remain highly capable but fundamentally jagged?
Today's models are superhuman on some tasks and fail at things a child can do — the 'jagged frontier.' Whether jaggedness is a transient artifact of training distributions or intrinsic to the paradigm is the crux.
View on the map → · Open in Browse →
Jaggedness stopped being an anecdote: DeepMind published a research program for measuring it, Epoch built an index betting a general capability factor sits underneath it, and the field-experiment tradition now argues the unevenness is structural to how models are trained rather than a passing phase. Mollick's revision is the one to read first: the frontier smooths in lurches, not uniformly.
Jaggedness creates bottlenecks, and bottlenecks mean that even very smart AI cannot easily substitute for humans.
The coiner of the 'jagged frontier' framing updates it: lab effort concentrates on 'reverse salients' — specific weaknesses whose resolution suddenly unlocks whole task areas — so the frontier smooths in lurches rather than uniformly.
Frontier AI models exhibit a paradoxical and uneven profile of competencies. They can achieve expert-level performance on many challenging tasks while failing at others that are simple for most people.
DeepMind's first systematic position paper treating jaggedness as a measurable property of models — moves the debate from anecdote to research program.
The frontier is jagged because it does not map to human intuitions about task complexity. It maps to something structural about how AI systems are trained.
Co-author of the canonical BCG jagged-frontier field experiment reviews four follow-up experiments, arguing jaggedness is structural to training rather than transient.
ECI is designed to capture a broad, underlying capability useful across many tasks, rather than performance on specific skills.
A composite index aggregating 50+ benchmarks onto a single capability scale — a direct empirical bet that a broad general-capability factor exists beneath jagged benchmark results.
Generative AI systems often display highly uneven performance across tasks that appear 'nearby'.
First formal economic model of jaggedness, treating uneven capability as an equilibrium property with implications for when users can and cannot rely on AI.