Archive · 15 Dec 2025
Human Typing Speed Limits AGI Progress
Alexander Embiricos, who works on OpenAI's Codex, says that the biggest problem in making super-smart AI (called AGI) isn't the technology itself.
It's how fast humans can type and check the AI's work.
Right now, AI is very fast, but humans can't keep up.
This slows down how quickly AI can improve.
Embiricos thinks the solution is to make AI systems that can check their own work without needing humans to help.
This could make everything much faster and more productive, like a big jump in how much we can get done.
He believes that once this happens, AI research will improve so much that we might finally reach AGI.
Human typing and multi-tasking speed are the main bottlenecks in AGI progress.
AI systems rely on humans to validate outputs, which slows innovation.
Embiricos proposes AI agents that can validate their own work to reduce human dependency.
This could lead to a 'hockey stick' growth in productivity, starting with early adopters next year.
Once productivity gains flow back into AI research, the conditions for AGI may emerge.
- Who
- Alexander Embiricos, OpenAI Codex lead
- What
- Human typing speed is the limiting factor in AGI progress
- When
- Predicted productivity gains starting next year
- Where
- Not specified, but likely in AI research and development
- Why
- Humans are not fast enough to keep up with AI systems, slowing feedback loops and innovation
- Key Figure
- Alexander Embiricos, OpenAI Codex lead
- Main Obstacle
- Human typing and multi-tasking speed
- Current Limitation
- Human validation of AI outputs
- Proposed Solution
- AI agents validating their own work
- Predicted Outcome
- Hockey stick growth in productivity
