What if AI worked at 1.000.000 tokens per seconds?
6 hours ago
- The thought experiment explores the implications of an AI writing one million tokens per second, emphasizing that speed does not eliminate the need for human judgment.
- Four distinct meanings of 'a million tokens a second' are clarified: context capacity, input processing, aggregate throughput, and single-agent output.
- Even at extreme speeds, tasks remain bottlenecked by steps like testing, verification, and human decision-making, as illustrated by Amdahl's law.
- Scarce resources at high speed include taste, clear specifications, physical evidence, and fidelity to real-world constraints.
- Practical advice includes defining 'done' with testable criteria, setting criteria before reviewing options, identifying slow steps, and seeking diverse assumptions.