Ethan Mollick’s experience with the Mythos AI, specifically the model Claude 5 Fable, provides a window into the future potential and the foundational changes AI could bring to human-computer interaction. His encounter with Fable indicates a paradigm shift where the AI not only aids but also increasingly takes charge of complex tasks traditionally managed by human experts.
Fable’s efficiency and efficacy are exemplified early in Mollick’s description. He relays how Fable consistently outperformed other public AI models in various tasks. Fable’s capacity to work autonomously for extended periods on multi-page specifications and to innovate in its responses was specifically highlighted. This capability was clearly demonstrated when Fable produced an academic social science paper of remarkable sophistication and a 10-page epic poem on command from simple prompts.
Additionally, Fable’s adeptness is showcased through lighter applications such as game creation. It successfully executed game development from vague prompts without relying on external assets. These examples serve to illustrate Fable’s considerable computational power and creative capability.
Despite these advantages, Mollick’s interactions with Fable led him to confront the unnerving efficiency of the AI — where commands were executed with disarming immediacy and accuracy, often with little to no ongoing input needed from him. This aspect was further explored during a more complex task: the creation of an isochrone map. Fable was instructed to create a detailed and aesthetically pleasing map showing travel distances from various cities, considering various modes of transportation. What differentiates Fable’s approach from previous AIs Mollick tested is its method of delegating subtasks to other AI models, refining its own outputs, and autonomously making design choices.
The creation of Concord, a sophisticated software aiming to calibrate human and AI responses for accurate data analysis, epitomizes Fable’s high capability. This project involved Fable generating a comprehensive design and then executing on it autonomously, showcasing an astounding degree of independent decision-making and application development.
Mollick frames his experience with a dual sense of marvel and disquiet. The AI’s power and independence are empowering yet unsettling. He describes the evolving relationship with AI as less about direct control and more about setting objectives and parameters and then assessing the outcomes. Thus, the human role tilts from ‘executor’ to ‘overseer’ or ‘patron,’ changing the nature of interaction and reliance on AI systems.
As Fable operates within strict guardrails that prevent usage for certain sensitive tasks (like cybersecurity), it often defaults to less powerful models in scenarios where security concerns are detected. Moreover, despite its profound capabilities, Fable’s operational costs—measured in computational ‘tokens’—are substantial relative to other models, which influences its practical deployment.
Mollick’s reflection touches on a broader, philosophical aspect of AI development. He ponders whether the increasing capabilities of AI will lead to a more opaque operational standard, where the processes and decisions made by AI become less accessible to human supervisors. The ‘black box’ nature of advanced AI could signify a shift not only in how tasks are completed but also in the fundamental understanding and transparency of decision processes in computational systems.
In essence, Mollick’s account suggests that as AI grows more adept at handling complex and creative tasks, it simultaneously becomes more autonomous and less interpretable. This could reshape the landscape of software development and other fields heavily reliant on AI, pushing forward the boundaries of what machines can achieve while also challenging our understanding of and interaction with these advanced systems.
Read the full post on oneusefulthing.org
