The Typing Speed Dilemma in AI Development

In a revealing podcast discussion, Alexander Embiricos, product lead at OpenAI Codex, highlighted an ironic challenge facing artificial intelligence advancement: humans simply type too slowly.

Image source note: The image is AI-generated, and the image licensing service is Midjourney.

The Unexpected Bottleneck

"We've built systems that can process information at lightning speed," Embiricos explained, "only to have them wait for human fingers to type out instructions." This manual prompting process, along with the need for human verification of AI outputs, creates what he calls "the last mile problem" of AGI development.

The issue goes beyond just typing speed. Even when AI agents can observe human workflows, they still require manual result checking - a time-consuming process that negates much of the efficiency gains from automation.

Rethinking the Human-AI Relationship

The proposed solution involves fundamental changes to how AI systems operate. "We need agents that work autonomously by default," Embiricos emphasized. "Instead of constantly asking for permission or waiting for prompts, they should be able to complete tasks independently unless specifically stopped."

This shift could trigger what economists call "hockey stick growth" - slow initial progress followed by sudden exponential acceleration. Early adopters might see their productivity skyrocket as they delegate entire workflows to AI rather than just individual tasks.

The Road Ahead

Full automation won't happen overnight. Each application area requires customized solutions, and tech giants will need time to implement system-wide changes. Embiricos predicts AGI will first make significant impacts in middle-layer applications before achieving complete autonomy.

The challenge now lies in developing new verification methods and system architectures that maintain quality control while reducing human oversight. As Embiricos put it: "We're not just improving AI - we're redesigning how humans and machines collaborate."

Key Points:

• Human typing speed has emerged as an unexpected bottleneck in AGI development
• Autonomous operation by default could unlock rapid acceleration in AI capabilities
• Early adopters may see dramatic productivity gains before full automation arrives
• System-wide changes will require new verification approaches and architectures