The author does an interesting prompting trick I hadn't considered trying with ChatGPT: ...and... It makes perfect sense, to allow an LLM workflow to self-improve by asking it to write its own...
The author does an interesting prompting trick I hadn't considered trying with ChatGPT:
What kind of information would have allowed you to avoid making those mistakes? I am not interested in a revised explanation, I want you to write the text that I could give to an LLM in addition to other information about Kłeti so that it wouldn’t repeat your mistakes.
...and...
Please summarize everything that you know about Kłeti’s grammar in one prompt that I could feed to ChatGPT so that it can translate between Kłeti and English, without making the mistakes you made.
It makes perfect sense, to allow an LLM workflow to self-improve by asking it to write its own prompts. Smart! I'm definitely going to adopt that approach myself.
That aside, I think it's awesome that his kid devised such an elaborate conlang. I dabbled in that sort of thing a bit when I was a kid, but never to such completion. I'm not surprised that they hit a wall getting ChatGPT to understand it fully but I'm quite impressed at how much it did pick up, and how quickly.
The author does an interesting prompting trick I hadn't considered trying with ChatGPT:
...and...
It makes perfect sense, to allow an LLM workflow to self-improve by asking it to write its own prompts. Smart! I'm definitely going to adopt that approach myself.
That aside, I think it's awesome that his kid devised such an elaborate conlang. I dabbled in that sort of thing a bit when I was a kid, but never to such completion. I'm not surprised that they hit a wall getting ChatGPT to understand it fully but I'm quite impressed at how much it did pick up, and how quickly.