Jagged AGI: Near‑AGI with Gaps

Large foundation models are reaching expert‑level performance across many computer‑based tasks (code, video, planning) but do so unevenly: they can ace some subtasks while failing others in non‑obvious, hard‑to‑predict ways. This 'jaggedness' is not simple hallucination but structural — competence cliffs across tasks and contexts that complicate deployment, auditing, and legal responsibility. — Policymakers, regulators, and organizations should treat advanced models as ‘jagged’ systems and design oversight, procurement, and labor policy around uneven failure modes rather than assuming either narrow or full AGI.