“The same experiments, the same figures, the same tool, all survive perfectly well as plain interpretability. Someone chose to wrap it in neuroscience. That choice is the product, not the science.”
RSRavid Shwartz Ziv@ziv_ravidTECH#741
Story Brief
In a long post on X, researcher Ravid Shwartz Ziv argued that Anthropic’s “consciousness” paper is being oversold. His point is narrower than a full technical takedown: he wrote that the underlying interpretability work and the J-lens tool look "fine" and "reasonable," but that Anthropic chose to frame standard model-analysis experiments with neuroscience language and the word "conscious" for narrative effect. In his telling, the math survives without the consciousness label.
Ravid Shwartz Ziv’s main claim is that Anthropic’s paper can be read as ordinary interpretability research without the consciousness framing.
The public argument is mostly about framing and marketing, not a claim that the underlying experiments never happened.
Visible X reactions split between readers who appreciated the critique’s nuance and others who saw the paper as another example of research being used to sell a bigger narrative.
Supportive reactions on X mostly praise the post for making a narrow distinction: the J-lens work may be useful even if the consciousness packaging feels loaded. Negative reactions on X are harsher and worry that Anthropic’s public research language pushes readers toward bigger claims than the technical results support. That visible mix says more about trust in lab messaging than about a settled verdict on the paper itself.
On X, much of the visible reaction backed that distinction. Replies on X called the critique well put and praised it for separating the paper’s technical method from the story wrapped around it. Sharper criticism went further than Shwartz Ziv’s own post, with some X users arguing that Anthropic blurs research and marketing when it talks about model behavior. The public posts show one researcher’s attributed argument about framing, plus a conversation less about whether the experiments exist than about what readers are supposed to think they mean.
Ravid Shwartz Ziv@ziv_ravidTECH#741
Flock49600209TECH
