> When you think about how people use ChatGPT, one of the brilliant things they did was, when we first started using it, it typed slowly. It didn’t just sit there waiting until it had the whole sentence figured out. It typed a few words, then it typed a few more words, then it typed a few more words. There was no technical reason for that. That was a marketing decision because it made us imagine that there was a little man or woman inside the thing typing back to us. They intentionally created personification.
This is 100% wrong -- that happens because the LLM generates the text token by token in real time. The alternative would be a very long loading animation, running until 100% of the output is done, and that would probably turn most people off.
This is 100% wrong -- that happens because the LLM generates the text token by token in real time. The alternative would be a very long loading animation, running until 100% of the output is done, and that would probably turn most people off.