Human behavior and cognition operate at 10 bits/second.
However, our brain processes sensory input at a rate of 1 trillion bits per second.
A typist typing at 120 words per minute is pretty much running at full capacity. Reading, recognizing, and initiating the motor skills of the hands to type each letter.
How Fast Is AI?
Because large language models like ChatGPT run on distributed GPU-based systems, we need to break them down in conceptual terms:
1. Token Throughput
OpenAI models like GPT-4 can process:
Input tokens: Up to ~128,000 tokens per prompt (for GPT-4-turbo).
Output speed: Around 20 to 100 tokens per second, depending on system load and context length.
Each token roughly equals four characters or ~3 bytes on average.
