LLM Counter-Defaults
"The tool I've been wanting but couldn't put my finger on it. Already adjusted my preferences, thank you!" - Brunni Corsato
"Thank you for sharing your work openly - and for challenging people to reflect on their own values with a practical tool." - Marlijn Gelsing, Kunstloc Brabant (translated from Dutch)
A small free tool that writes the custom instructions to override your LLM’s defaults. ChatGPT, Claude, and Gemini ship tuned to agree with you, sound certain even when they’re wrong, and write in your place. Counter-Defaults lets you set your own rules: you move sliders across 14 behaviors – how much the model pushes back, flags what it can’t verify, scaffolds your reasoning instead of handing you an answer, protects your writing voice – and paste the result once.
Launched June 2026 · free, open, no sign-up · works with any LLM that has a custom-instructions field → counterdefaults.aixdesign.co
From my launch note: “I use LLMs daily, but I don’t want to give in to cognitive surrender, find my own thinking smoothed into mid, and waste the energy and water it takes to run any prompt on flattery and filler.” The tool is grounded in research on sycophancy, homogenized output, and the deskilling that creeps in when we adopt LLM answers even when they’re wrong.
I built it, with testing and nudging from the AIxDESIGN community, using Claude Code. It has since traveled into the classroom: a workshop with arts educators for Kunstloc Brabant, where the sliders become a way to talk about your values as much as your settings. More sessions are planned.
