Reason #23,543 why I won't be using an LLM to write. When you crowd source training one of those things, you get a monstrosity spitting out trash output. GIGO.
And, God forbid, we end up with a well trained monstrosity, it will spit out mediocre work at such a high rate that no one can compete in an ever more saturated story market.
Thanks for digging through the trash to find if there were any pearls in ChatGPT, Ms. Monalisa. 🙇♂️
This is a fairly odd test designed for the AI to fail. Upload your book to it and get it to do the same thing. Claude Pro and Sudowrite both do a really good job creating a synopsis from an uploaded book and can provide you a lot of information once it is kept in memory. Sudowrite even managed to pickup on something that was not explicitly mentioned in my book, but was in my notes (that I had not given it)
Asking it to go out and research something is not an odd test at all. That is the purpose of the Deep Research function. I did not expect it to make up a bunch of stuff. And I have had it summarize chapters and scenes. It still hallucinates so it’s not an unfair test at all. It shows how it’s generative even when summarizing. It’s not designed to admit that it doesn’t know something.
Huh. That's interesting and creepy in equal measure. Thanks for the info and insights. I skirt around the edges of a writing-based AI program, using it for punctuation and grammar. Not in too deep.
Thanks for "taking one for the team", in all the research you do on these subjects. There are very few things I trust an LLM or gAI with, just like my dogs or my kids when they were toddlers. (And "Trust but Verify" should still be part of the SOP, even then).
Reason #23,543 why I won't be using an LLM to write. When you crowd source training one of those things, you get a monstrosity spitting out trash output. GIGO.
And, God forbid, we end up with a well trained monstrosity, it will spit out mediocre work at such a high rate that no one can compete in an ever more saturated story market.
Thanks for digging through the trash to find if there were any pearls in ChatGPT, Ms. Monalisa. 🙇♂️
This is a fairly odd test designed for the AI to fail. Upload your book to it and get it to do the same thing. Claude Pro and Sudowrite both do a really good job creating a synopsis from an uploaded book and can provide you a lot of information once it is kept in memory. Sudowrite even managed to pickup on something that was not explicitly mentioned in my book, but was in my notes (that I had not given it)
Asking it to go out and research something is not an odd test at all. That is the purpose of the Deep Research function. I did not expect it to make up a bunch of stuff. And I have had it summarize chapters and scenes. It still hallucinates so it’s not an unfair test at all. It shows how it’s generative even when summarizing. It’s not designed to admit that it doesn’t know something.
Huh. That's interesting and creepy in equal measure. Thanks for the info and insights. I skirt around the edges of a writing-based AI program, using it for punctuation and grammar. Not in too deep.
Thanks for "taking one for the team", in all the research you do on these subjects. There are very few things I trust an LLM or gAI with, just like my dogs or my kids when they were toddlers. (And "Trust but Verify" should still be part of the SOP, even then).
Or maybe it was sandbagging your watchful eye!