A Bayesian essay puts frontier LLM functional self-awareness at 75% and phenomenal awareness at 12%

Individual-Advice215 · reddit · 2026-07-24

This long post asks how likely it is that a frontier LLM is self-aware, and explicitly separates two notions:

Functional self-awareness: the model can represent and use information about its own role, uncertainty, intentions, or reasoning state.
Phenomenal self-awareness: the model has subjective experience—something it is like to be the model.

The author says the estimate is Bayesian rather than experimental, combining interpretability evidence, signs of internal self-monitoring, limitations of introspection, differences from biological brains, and unresolved consciousness theory.

Key conclusions:

Functional self-awareness: 75%
Suggested range: 61–89%
Phenomenal self-awareness: 12%
Suggested range: 3–21%
Persistent autobiographical self between sessions: probably below 5%

Why the gap?

Functional self-awareness has observable markers: frontier models can form internal representations that are reportable, reusable, and causally involved in reasoning.
Phenomenal self-awareness is much harder to infer; internal self-monitoring does not prove subjective experience.
The author gives little weight to the model’s own claims of being conscious, since those are heavily shaped by training and prompting.

Bottom line: the post argues frontier LLMs probably have a narrow, temporary, unstable form of functional self-awareness, while subjective awareness remains far less likely.

Original post →

A Bayesian essay puts frontier LLM functional self-awareness at 75% and phenomenal awareness at 12%

More from AGI Musings