Should we extend test model to what university professors can teach?
Should we extend that model to text books? If I learn about a topic from a book, can I never write a book of my own on that topic?
Should we extend that model to the web? If I learned CSS and JavaScript reading StackOverflow, am I banned from writing a book, giving classes, or indeed even answering questions on those topics?
I ask this in seriousness. I get that LLM training is new, and it's causing concerns. But those concerns have existed forever - the dissemination of information has been going on a long time.
I'm sure the same moral panic existed the first time someone started making marks in clay to describe what us the best time of year to plant the crops.
No, because none of those things involve creating a copy on a computer which will then be regurgitated w/o acknowledgement of what went before, and w/o any sort of compensation to the previous rights bearer.
Time was if a person read multiple books to write a new text, they either purchased them, or borrowed them from a library which had purchased them, and then acknowledged them in a footnote or reference section.
At least one author noted that there was a concern that writing would lead to a diminishment of human memory/loss of oral tradition (Louis L'Amour in _The Walking Drum_).
> At least one author noted that there was a concern that writing would lead to a diminishment of human memory/loss of oral tradition (Louis L'Amour in _The Walking Drum_).
Really? I can't tell if you're joking, so I'll take it at face value.
See, I associate the earliest famous (I thought) expression of that concern with Plato, and before today I couldn't remember any other associated details enough to articulate them with confidence. ChatGPT tells me, using the above quote without the citation as a prompt, that it was in Plato in his dialogue Phaedrus, and offers additional succinct contextual information and a better quote from that work. I probably first learned to associate that complaint about writing with Plato in college, and probably got it from C.D.C. Reeve, who was a philosophy professor and expert on Plato at the college I attended. But I feel no need to cite any of Reeve's works when dropping that vague reference. If I were to use any of Reeve's original thoughts related to analysis of Plato, then a reference would be merited.
It seems to me that there are different layers of abstraction of knowledge and memory, and LLMs mostly capture and very effectively synthesize knowledge at layers of abstraction that are above that of grammar checkers and below that of plagiarism in most cases. It's true that it is the nature of many of today's biggest transformers that they do in some cases produce output that qualifies as plagiarism by conventional standards. Every instance of that plagiarism is problematic, and should be a primary focus of innovation going forward. But in this conversation no one seems to acknowledge that the bar has been moved. The machine looked upon the library, and produced some output, therefore we should assume it is all theft? I am not persuaded.
Should we extend that model to text books? If I learn about a topic from a book, can I never write a book of my own on that topic?
Should we extend that model to the web? If I learned CSS and JavaScript reading StackOverflow, am I banned from writing a book, giving classes, or indeed even answering questions on those topics?
I ask this in seriousness. I get that LLM training is new, and it's causing concerns. But those concerns have existed forever - the dissemination of information has been going on a long time.
I'm sure the same moral panic existed the first time someone started making marks in clay to describe what us the best time of year to plant the crops.