The training data set is a huge corpus of text from a wide variety of styles and time period. A human writer knows to adjust their writing style to fit their intended audience. A skilled LLM user knows to direct the LLM to generate in the appropriate style. A naive LLM user doesn't, so that's how we get this uncanny valley ick, where the text feels wrong.
The training data set is a huge corpus of text from a wide variety of styles and time period. A human writer knows to adjust their writing style to fit their intended audience. A skilled LLM user knows to direct the LLM to generate in the appropriate style. A naive LLM user doesn't, so that's how we get this uncanny valley ick, where the text feels wrong.
It was always corny except in very specific cases.
The network is the computer
https://en.wikipedia.org/wiki/The_Network_is_the_Computer
All the marketing materials that got used for training. Marketing speak is not my cup of team.
It has been used before LLMs. I have no idea why LLMs prefer it, though.