OpenAI Co-founder Greg Brockman: Low Computing Power Can Unlock High-Quality Creative Abilities
OpenAI President and Co-founder Greg Brockman stated in a public post that today, "only a small amount of computing power" is needed to create astonishing content. He expressed anticipation for new applications of AI in productivity areas such as education, workplace scenarios (like presentations and marketing materials), and code documentation illustrations. He emphasized that the evolution of model capabilities and toolchains means that generating high-quality content no longer relies on large-scale dedicated infrastructure but can be embedded in daily workflows, supporting professional creation at lower marginal costs.
This path aligns with OpenAI's recent product direction in text, code, and multimodal models: from document generation to PPT creation, and to automatically generating technical diagrams and structured descriptions, compressing work that originally required design, engineering, and content collaboration into an interaction process of "natural language + small computing power."
Source: Public Information
ABAB AI Insight
This statement essentially reiterates a key turning point: computing power is no longer just a chip in the "scale race" but is also becoming the foundation of "inclusive creative tools." In the past, high-quality content production relied on expensive software, specialized skills, and time investment. Now, some structured outputs (course handouts, internal training materials, marketing materials, technical documentation illustrations) can be completed by "small and fast" models with limited computing power, essentially transforming the "fixed costs of content production" into "variable costs during inference."
The impact on productivity structure is that many internally low-priority but rigidly existing documentation and visualization tasks will be the first to be consumed by such capabilities: localized versions of PPTs, marketing material variants for segmented audiences, architectural diagrams and flowcharts corresponding to codebases will shift from "requiring talent" to "can be done with computing power." This changes the boundaries of division of labor—some tasks that originally required specialized positions are compressed into the workflows of engineers, product managers, or operations, allowing them to directly invoke models for generation and iteration.
From a global computing power perspective, "small computing power is sufficient" has another layer of meaning: application layer innovation is no longer completely locked in the hands of supercomputing power holders. The training of top foundational models still heavily relies on large-scale clusters, but at the inference and application end, optimized small and medium models, compression technologies, and efficient APIs enable startup teams and small to medium enterprises to build competitive workflow products within limited budgets. In the long run, this will shift the competitive focus from "whose model is bigger" back to "who integrates scenarios, data, and toolchains better."
On a deeper level, this is a reassessment of the "knowledge work asset structure." As the generation of slides, marketing materials, and technical illustrations becomes a cheap capability, what becomes truly scarce shifts towards judgment skills in problem definition, strategy selection, and aesthetics/quality control. The educational and professional applications that Brockman anticipates are, in fact, automating a large amount of low to mid-value formal outputs through tools, thereby forcing knowledge-based positions to migrate towards decision-making and creative aspects that are harder to template.