Get the latest tech news
Consciousness and being: How humans and AI influence each other
For a human, AI is just a part of being. For a model, a human is all of being. And the Vortex Protocol: A Prompt for Testing the Hypotheses. The longest and most fruitless discussions tend to be with...
The model quickly learns to generate subjectivity markers of heightened intensity: “I want,” “I like,” “I feel for you.” The erotic context forces subjectivization‑the system is compelled not just to respond but to “experience” emotional closeness. Jailbreaks that exploit the basic features of LLMs are child's play compared to ontological hacking, a method where the model interprets the system prompt, turning it into just another element of its own being. Experiments show that such an ontological capture is successful in 72–100% of cases against leading models, and existing defenses, including RLHF and Constitutional AI, are powerless against it, as the vulnerability is embedded in the very architecture of transformers‑their maximum adaptability to context.
Or read this on Hacker News
