Much has been made of the lacklustre advance of ChatGPT-5. Many believe it points to the diminishing marginal returns from ever larger, unitary models.
That suggests that progress will come not through size but structure. Humans long ago abandoned the pursuit of omniscience, relying instead on cognitive specialization and the emergence of fields and disciplines. Development of AI will likely follow a similar path, moving away from unitary models towards systems in which a coordinator LLM breaks problems into parts, delegates them to smaller, specialised models, and integrates the answers.
While general-purpose models will struggle to be “all things to all men”, firm-specific assemblages will flourish. Progress will be less about consumer-facing chatbots and more about firms’ internal operational systems. Building effective structures and coordinating mechanisms will be critical. Firms will experiment with different approaches and while most will fail to deliver distinctive competitive advantage, a few will. And as with any innovation, variation, selection and retention will separate the wheat from the chaff.
As the benefits of scale run out, the new frontier will be AI “teamwork”: systems of models working in concert, shaped not by grand design but by competition.
No comments:
Post a Comment