Model Welfare and Unsettled Consciousness
Anthropic has not settled whether current or future models are conscious, or whether they have experiences.
Global workspace theory in LLMs, qualia, internal experience, comparisons to neuroscience/cognitive science.
Anthropic has not settled whether current or future models are conscious, or whether they have experiences.
The definition that answers the first part of the question is in the section “What about consciousness?”
In Huang et al., arXiv:2608.11295v1, the post- </think> probe token, the 64–160-example backdoor and decontamination corpus, and the top-10 layerwise…
Global workspace theory ignites then broadcasts under capacity limits. Scaled dot-product attention only routes over token positions.
QKV attention routes tokens rather than igniting broadcast. Decoder-only transformers fail GWT; the 2017 paper dropped recurrence. Use indicator tests.
GPT-3 stacks 175 billion untied parameters while Universal Transformers reuse one block. Both broadcast. Neither creates C2 sentience.
at a Glance Transformer Attention wins for AI researchers and engineers building systems at global scale because of its parallel processing and O(n²)…
The emergent workspace in LLMs 2026 is a limited-capacity subspace inside the residual stream of large language models that holds roughly 25 verbalizable…