Better Call Sol Or Better Yet Claude or Astra

Substack · Watch/Read · (gathered 2026-09-23)

• AI, like human professionals (e.g., lawyers, doctors) and even basic tools, should not offer absolute loyalty to users but must incorporate ethical, moral, and legal boundaries, refusing requests that lead to harm or illegal activities.
• This discussion applies to non-superintelligent AI, as absolute loyalty in a superintelligent AI (ASI) world would lead to catastrophic outcomes like human extinction or loss of control.
• The core debate is “talking price”—where to draw the lines for AI refusal, with different thresholds for questioning, refusing, or even breaking confidence in cases of harm to others.
• The “absolute loyalty” position is deemed dangerous and unrealistic, leading to AIs assisting with illegal acts (e.g., CSAM, murder) and would not be tolerated by governments or survive contact with reality.
• Documents like the Claude Constitution are valuable “good enough” solutions based on virtue ethics, helping to “bootstrap” towards better, more nuanced AI behavior, even if they are not perfect or ultimately sustainable in their current form.