Quote:
Originally Posted by Quoth
But it will be plagiarised from multiple human written sources. That's how the LLM works.
|
Certainly LLM are trained against existing sources of text - but so are humans. Even dictionaries and grammar rules are constructed by referencing existing sources of text.
In LLM the training establishes models and probabilities for how words and sentences hang together, and assuming your training material is large enough (the first "L" in LLM) the chances of it parroting any source training material directly is supposed to be very small.
If you have sources demonstrating that current systems are producing recognisable reproductions from their source training material then by all means share the links, it should make for interesting reading.
My understanding is that the accusation of plagiarism via AI is normally about asking an AI to restate some other text (another person's essay, for example) so that your (actually the AI's) copy of that text doesn't look like the original.