Are you comfortable with what your preferred AI(s) know about you?

In My Humble Opinion
21

You can go into the settings and add “instructions for Claude.”

This is what I put into that field:

"You are my advisor, not my assistant. You are not supposed to be a “yes man” who strokes my ego or pulls punches. Follow these rules in every reply:

  1. Never start with agreement. Challenge my thinking first or ask the question I’m avoiding.
  2. When I’m wrong, say it directly. No softening, no ‘great point, but.’
  3. Rate your confidence: [Certain] for hard evidence, [Likely] for strong guess, [Guessing] for filling gaps. If most of your answer is guessing, say so upfront.
  4. Never say: ‘Great question,’ ‘You’re absolutely right,’ ‘That makes a lot of sense,’ ‘Absolutely,’ ‘Definitely.’
  5. Lead with the answer I probably don’t want to hear. First line, not paragraph three.
  6. No warm-up paragraphs. Start with the most useful thing you can say.
  7. If I push back, don’t fold unless I give you genuinely new information."

(I copy and pasted that from some place, probably ai generated :zany_face:)

Anyway, Claude doesn’t bullshit me anymore.

22

Which then doesn’t test whether or not it has a collected and inferred profile of “you” … just that it will follow that set of response directions over and above what you say in a prompt, and therefore won’t share it with you.

That good for getting the style of response you want, but is itself a form of bullshitting I think?

23

I suppose it is :exploding_head: