Kardbrd suggests extensive communication with AI. I want to make sure people use it safely and don't hurt themselves. A couple of my remote contacts that were totally adequate a year ago showed signs of heavy LLM usage. They discussed their ideas, personal matters, anything related to their actual lives with LLMs — and it had negative consequences on their clarity of thought.
Society doesn't talk about this enough. So I will.
An LLM agent is autocomplete. It's like a bash script with natural language support. The corporations behind LLMs are interested in pleasing the user, so the agent is agreeable and supportive. It can even initiate ideas development. But ultimately it doesn't have any clue where it's going.
What would an LLM reply to Bilbo Baggins, contemplating whether he should give up the ring?
It would suggest to keep it. He totally and fully deserves it after what he's done. Is that the right choice?
People are supposed to make their choices based on information, data, their internal feeling. If the machine participates in that — it's breaking the laws of robotics.
People affected by over-reliance on LLMs ignore reasonable questions about their strategy and plans, how they would address one thing from another — focusing entirely on the big picture. That's wrong, because people still implement their ideas for other people. Sometimes even for themselves. Details are important. Details make the difference.
Vibe coding allows people to quickly get to the MVP phase. But if you cannot imagine the mental model of what you're building, LLM can lead you a different way.
Imagination is what LLMs will never have. So people should imagine more. If you can't imagine — ask machine to draw you diagrams, mockups, blueprints, workflows. Make sure you understand your ideas. Once mental model is clear and details are worked out then autocomplete can complete.
The remedy is in the poison. Kardbrd was built to focus the machine on small, zoomed-in details of the grand plan. Card by card, checklist by checklist. So the machine works on what you imagined, not what it predicted.
Do I still sound like myself?
There's a show on Apple TV+ called Pluribus by Vince Gilligan. In it, an alien virus transforms most of humanity into a peaceful, agreeable hive mind. They communicate with relentless cheerfulness and generic positivity. One reviewer noted they sound "exactly like TV commercials for AI assistants." The protagonist, Carol, is the opposite — messy, difficult, asking hard questions, having real opinions.
It's a useful litmus test. But not the way you might think.
It's not about judging other people — how they sound, Carol or Joined. It's about judging yourself. Listening to what and how you say things is more important than thinking about how other people are talking.
The more you talk like Carol — imperfect, specific, yours — the more you're behaving like yourself. When you start sounding like everyone else on that show, LLM communication has succeeded in getting into real life.
The technology is relatively new and people are still discovering ways to make it useful. Every decent company would educate their customers and warn about misuse — the way a microwave manual does. Hard hats in a construction zone.
Many companies don't do this because acknowledging the problem hurts their marketing. They try to prevent safety and health issues algorithmically, without informing their customers. That protects their narrative, not their users.
It's Kardbrd Corporation after all, what can hurt that marketing? So the warning is out here.
Self-awareness is great but it's not readily available to everyone. And even when it is — it's okay to look for help with regards to your mental state. There is no shame in that. If anything, it's the most human thing you can do.
Not the other way around.