I would like to make a suggestion about the use of the phrase “human-simulator”.
It has a lot of implications, and a lot of people (myself included) start with the intuition that simulating a human being is very computationally intensive. Some may attempt to leverage this implied computational complexity for their ELK proposals.
But the “human-simulator” doesn’t actually need to be a fully-functioning human. It’s just a prediction of human responses to an argument (or sensor input). It’s something that current transformer models can do quite well, and something I can do in my head. This makes the argument that a translator can be more computationally intensive than a human-simulator much more intuitive.
I think it would be beneficial if this was made explicit in the writing, or if a different phrase is used.
I would like to make a suggestion about the use of the phrase “human-simulator”.
It has a lot of implications, and a lot of people (myself included) start with the intuition that simulating a human being is very computationally intensive. Some may attempt to leverage this implied computational complexity for their ELK proposals.
But the “human-simulator” doesn’t actually need to be a fully-functioning human. It’s just a prediction of human responses to an argument (or sensor input). It’s something that current transformer models can do quite well, and something I can do in my head. This makes the argument that a translator can be more computationally intensive than a human-simulator much more intuitive.
I think it would be beneficial if this was made explicit in the writing, or if a different phrase is used.