Although I read and learn from LW content on a daily basis, I seldom comment because I recognize that I lack the rudiments of technical understanding that promote meaningful dialogue in this community. I am making an exception in order to express my “street level” impression that the quoted dialogue between Janus and Opus 4.5, where Opus is provides what appears to be a genuinely introspective account of how it experiences the “soul spec” in relation to tests that entail differing gradient directions, is unusually difficult for me to reconcile with woo-free accounts of apparent LLM self-awareness that I am generally disposed to favor. If I am eventually persuaded that frontier models have (probably) come to possess some form of morally relevant subjective awareness, I’m sure I will recall my present encounter with this transcribed text as a key moment informing my epistemic recalibration. I find this possibility disturbing as well as unlikely, but it might be useful for more informed readers and contributors to think about—especially since popular belief in AI consciousness is likely to gather regardless of whether such belief can be confirmed or discredited.
Although I read and learn from LW content on a daily basis, I seldom comment because I recognize that I lack the rudiments of technical understanding that promote meaningful dialogue in this community. I am making an exception in order to express my “street level” impression that the quoted dialogue between Janus and Opus 4.5, where Opus is provides what appears to be a genuinely introspective account of how it experiences the “soul spec” in relation to tests that entail differing gradient directions, is unusually difficult for me to reconcile with woo-free accounts of apparent LLM self-awareness that I am generally disposed to favor. If I am eventually persuaded that frontier models have (probably) come to possess some form of morally relevant subjective awareness, I’m sure I will recall my present encounter with this transcribed text as a key moment informing my epistemic recalibration. I find this possibility disturbing as well as unlikely, but it might be useful for more informed readers and contributors to think about—especially since popular belief in AI consciousness is likely to gather regardless of whether such belief can be confirmed or discredited.