Fable 5.2 and Jagged AGI

Claude Fable 5.2 is presented as “almost” artificial general intelligence under a limited definition: performance at the level of most human experts across most computer-based intellectual tasks. Its abilities now extend well beyond coding, but remain uneven in ways that are subtler than ordinary mistakes. In demonstrations, it turned an obscure programmable keyboard into a set of games, devised a plausible business based on public records, and reconstructed W. H. Auden’s demanding 1941 literature course through extensive research and a polished educational site. These projects show strong initiative, research, design, and technical execution, often compressing months of work into hours.

The weaknesses emerge in judgment, reliability, and autonomous follow-through. The business plan was credible but unproven, and its systems for operating the venture were incomplete. The Auden site began accurate but dull; Fable needed human direction to improve its design and literary interpretations. Its entrepreneurship study found that lower company-registration fees increased registrations without comparable gains in later business activity, but it failed to preregister properly, produced an unwieldy paper, and still needed human verification. Other projects - including analysis of 1,186 stories and a growing, AI-built 3D world with hundreds of scenes - show striking creative range. Yet the sheer volume of generated work can exceed people’s capacity to check it. The remaining gap is not simply whether AI can produce impressive results, but whether it can consistently recognize what is good, verify its work, and manage complex tasks without human oversight.