Writing without AI

harr

@andang76 said:
So, I tried to find criteria that would allow me to determine what is right or wrong:

• Do I want the skill required for this task to be continuously exercised?
• Is the operation I am delegating to the AI an exercise in deep thinking? If so, I am giving up an exercise of great cognitive value.

Other criteria I like:

• Can I get the same or better results with other tools? For example with a search engine, encyclopedia or dictionary?
• Can I strengthen my memory by trying to remember an information first?
• Can I use this opportunity to improve my notes by checking my notes first?

@andang76 said:
I still have to evaluate whether breaking down someone else's text that I don't fully understand is something I should do myself or if I can leave it to the AI.

I do both. Comparing my own initial analysis with AI's initial analysis produces the best results. I also like to challenge AI's answer when it misses or misinterprets something. AI can be valuable, especially for longer texts.

I don't delegate my thinking to AI. I don't let AI decide for me, what is important to me.

Marcel_Liauw

@harr said:
(...) Comparing my own initial analysis with AI's initial analysis produces the best results. I also like to challenge AI's answer when it misses or misinterprets something. AI can be valuable, especially for longer texts.

Thank you for sharing. I frequently seem to hear that people obtain very different results from different AIs. Claude is better for this than Perplexity, ChatGPT better for that than Gemini. Which would be your favorites for the zettelkasten?

tomp

@Marcel_Liauw said:

@harr said:
(...) Comparing my own initial analysis with AI's initial analysis produces the best results. I also like to challenge AI's answer when it misses or misinterprets something. AI can be valuable, especially for longer texts.

Thank you for sharing. I frequently seem to hear that people obtain very different results from different AIs. Claude is better for this than Perplexity, ChatGPT better for that than Gemini. Which would be your favorites for the zettelkasten?

I use CoPilot for discussing things. It's very wordy but I'm not paying for the tokens and I can reduce the wordiness. It also likes to pick up what it thinks are patterns in the conversations, invent jargon for them, and insist on using it. I can tone down these tendencies with prompts.

The fact that it is somewhat unimaginative and stereotyped is not necessarily a disadvantage because with correcting the chatbot comes the most valuable mental work, just as it does when talking with a person.

Any bits (always my own writing) that seem cogent enough I've been pasting into an open text file to turn into z-cards at some point when I feel the material has matured enough.

harr

@Marcel_Liauw said:
I frequently seem to hear that people obtain very different results from different AIs. Claude is better for this than Perplexity, ChatGPT better for that than Gemini. Which would be your favorites for the zettelkasten?

I have tried all of them and more. :-)

Currently I'm very happy with Proton's Lumo as online chatbot. It's privacy conscious and uses a custom mix of open models. It has just the right amount of features for my needs.

Perplexity Pro used to be my favorite chatbot for research, but since they introduced their "Computer" feature, they try to push unnecessary tasks that consume way too many tokens for an absurd amount of money.

On my Mac I run Gemma and Qwen locally with decent results. They are good enough for my early experiments with agentic workflows. I prefer local AI because it's more private (no data leaves the computer) and because it costs less money (no per-token fee).

I'm still figuring out AI for Zettelkasten. ;-) So far the limiting factor isn't the model, but data access, search tools and prompting skills.