finding

active

finding:llm-optimist-s-assessment-falls-to-0-099-when-credence-is-concentrated-on-levels-4-5

LLM-optimist's assessment falls to 0.099 when credence is concentrated on Levels 4-5

Demonstrates that theoretical credence weighting can drastically lower the assessed probability for the same evidence.

Source paper

extracted_from

From cacophony to hierarchy: a principled framework for assessing AI consciousness

(2026) · Shamil Chandaria · Arvo Muñoz Morán · Fernando Rosas · Anil Seth +10

Neighborhood — ranked by edge-count

Papers (1)

paper

From cacophony to hierarchy: a principled framework for assessing AI consciousness
mentions

Claims (1)

claim

For current LLMs, the overall credence in consciousness is driven as much by where theoretical credence is placed across levels as by how the evidence is read
supports
Key finding of the Bayesian illustrative analysis in Section 7.4.7.

Frameworks (1)

framework

Bayesian Network Model of Consciousness Attribution
associated_with
The paper's formal Bayesian machinery translating the supervenience hierarchy into conditional independence and aggregated credence.

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

I think it wouldn’t be unreasonable to have a credence over 50 percent that we’ll have sophisticated LLM+ systems ... within a decade. It also wouldn’t be unreasonable to have at least a 50 percent credence that if we develop sophisticated systems with all of these properties, they will be conscious.quote0.768
David Chalmers' estimate of AI consciousness probability, quoted in §2.2.2.
LLM judge (GPT-4.1-mini) achieves 94.7% agreement with human judges across 300 pairwise comparisons for evil, sycophancy, hallucinationfinding0.765
Validates the automated trait expression scoring pipeline
Five independent LLM scorers from four labs produce identical rankings (Spearman ρ > 0.8).finding0.764
Scorer bias validation: Claude Haiku, Gemini Flash, GPT-5.4, Grok 4, Kimi K2.5 all converge on same model ordering.
Kendall's τ = 0.76 (p<.001) for Conscientiousness dimension LLM scoring vs human judgmentfinding0.762
Validates GPT-4o scoring reliability for Conscientiousness personality dimension
The Digital Consciousness Model finds a median posterior of ~0.08 (from prior ~0.17) that 2024 LLMs are conscious, varying substantially across theoretical stancesfinding0.758
External replication-adjacent result cited as consistent with this paper's level-structure prediction.
Clear accuracy stratification across three reflection levels on cruxeval_o_adv: Triggered (.065/.247) > Intrinsic (.040/.133) > No Reflection (.017/.051) for Qwen2.5-3B/Gemma3-4B-ITfinding0.756
Core empirical result validating the three-level reflection framework on code reasoning.
The better an LLM is at language modeling, the more it aligns with vision models, and vice versa — linear relationship between language modeling score and vision-language alignmentfinding0.752
Core cross-modal empirical result: larger and better language models align better with vision models
Better LLMs (measured by 1-bits-per-byte on OpenWebText) show a linear relationship with alignment to vision models measured via mutual nearest-neighbor on WITfinding0.747
Key cross-modal alignment result