This is an experiment in using a markdown architecture coordinated through llms.txt to make a book and associated content as legible as possible to LLMs
It's an issue! As someone who's worked on science policy for decades, including time in the Beltway, you cannot develop a robust and world-leading science policy strategy with technologists alone!
I find it depends on context. I write a lot as an academic and author. When I need to generate functional content that has a specific purpose (knowledge base, transfer of information etc) I will use AI where it makes sense. Where I write to explore ideas, develop my own thinking, and connect with others in a very relational way, I intentionally do not use AI. Plus, when I do this, writing is an extension of my identity and I'd rather not give that away!
I suspect the lack of consequences to getting mad help - no relationships that can be broken and that need mending. Not sure it's healthy, and yes, I do it when Claude simply does not get it or I know i could do better :)
My experience is that it all comes down to personal fit and feel. I switched from ChatGPT to Clause several months ago and much prefer it - although do get frustrated at glitches and hitting limits. But I'm a writer and academic, and the LLM fits my purpose better. With what I do ChatGPT does nott feel great to use.
Looks like a useful tool - just beginning to explore adding llms.txt to my sites. Frustrated that because it isn't a standard many LLMs will not look at or use, but a potentially low friction step toward future proofing!
The original files can be browsed on GitHub: https://github.com/2020science/spoileralert-wtf