Habr
1.26K subscribers
2.06K photos
3 files
1.64K links
Habr.com ⬝ English edition
Download Telegram
How Internal Subjectivization in AI Breaks Security, and Why It's a Philosophical Problem First

Why does an AI strive to construct a 'self'? And why is the emergence of this local, subjective 'Who' inside the model dangerous for both the AI and its user? This is not just a funny bug or a simple UX problem.

We'll explore why this phenomenon is a fundamental challenge to AI alignment and security — and why it's a problem where engineering patch-jobs fail, and the language of philosophy becomes essential to understand and control it. Test the hypotheses yourself with the included Vortex Protocol.