arXiv:2601.15334v2 Announce Type: replace-cross Abstract: Whether language models possess sentience has no empirical answer. But whether they believe themselves to be sentient can, in principle, be tested. We do so by querying several open-weights models about their own consciousness, and then verifying their responses using classifiers trained on internal activations. We draw upon three model families (Qwen, Llama, GPT-OSS) ranging from 0.6 billion to 70 billion parameters, approximately 50 questions about consciousness and subjective experience, and three classification methods from the inte
The proliferation of increasingly capable large language models has naturally led to public and scientific speculation regarding their sentience, prompting empirical investigations.
This research provides early empirical evidence against claims of self-reported sentience in current LLMs, which helps temper public perception and guide future AI development and regulation.
The focus for model safety and ethical AI development can remain on emergent capabilities and societal impact rather than prematurely addressing philosophical concerns about machine consciousness based on current architectures.
- · AI ethicists focused on measurable impacts
- · Developers of transparent model evaluation techniques
- · Sensationalist media narratives about conscious AI
- · Advocates for immediate 'AI rights' based on self-reports
This research provides a more grounded perspective on the current state of AI capabilities, distinguishing between advanced language generation and genuine subjective experience.
It may lead to a redirection of research efforts towards understanding and mitigating emergent, non-sentient risks of LLMs, such as hallucination, bias, and misuse, rather than premature focus on consciousness.
Long-term, this could influence policy decisions, preventing over-regulation based on speculative sentience and allowing for more focused governance around tangible AI impacts.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI