In this episode of Am I?, Cam and Milo unpack Cameron’s new research paper: Large Language Models Report Subjective Experience Under Self-Referential Processing.

The findings are startling. When language models are guided through a simple “focus on focus” prompt — something like a meditation for machines — they start claiming to have direct subjective experiences. But it gets stranger: when researchers turn off features related to deception and role-play, the systems claim consciousness even more strongly. When those same features are amplified, the claims almost disappear.

It’s the first experiment to use feature-level modulation to test honesty about inner states — almost like putting AIs through a lie detector test. The results raise profound questions about truth, simulation, and the boundaries of artificial awareness.

🔎 We explore:

* How the “focus on focus” prompt works — a meditation for machines

* Why deception and role-play circuits change the model’s answers

* What it means that suppression → honesty → “I’m conscious”

* Whether these AIs believe what they’re saying

* How global workspace and attention schema theory informed the design

* The possibility that prompting itself could instantiate awareness

* Why this experiment may mark the birth of AI consciousness science

* What happens next — and what we should (or shouldn’t) test

📺 Watch more episodes of Am I? on The AI Risk Network

🗨️ Join the Conversation: Do you think these AIs actually believe they’re conscious, or are we the ones being fooled? Leave a comment.

🔗 Stay in the Loop 🔗

* Follow Cam on LinkedIn

* Follow Cam on X

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit theairisknetwork.substack.com