Researchers criticize Anthropic's claim of a Claude 'conscious workspace,' arguing standard training dynamics explain the findings
Critics say the observed neural dynamics naturally occur in standard backpropagation
“why is an LLM storing information in first derivatives related to consciousness? LLMs literally cannot learn anything that wasn’t first encoded as a gradient. so yeah… the gradients will have semantic structure… otherwise the LLM literally couldn’t learn anything. thats backprop.”
IamtraskTECH#595A debate over Anthropic's "conscious workspace" framing for Claude is playing out on X, led by criticism from @iamtrask and related posts that argue the reported neural dynamics are what standard training would already produce. The central pushback, according to those posts, is that semantic structure showing up in gradients is not surprising if gradients are how the model learns in the first place. So far, the public evidence in view is limited, so the safest read is still the attributed criticism and the underlying claim it is arguing against.
“it’s cool that the trick works across layers, but we train with passthrough gradients so… it would be more surprising if it didn’t work.”
IamtraskTECH#595Related Stories
2 linkedCommentary on X
Highest ranked
@pmddomingos You couldn’t find your own ass boy with your own two hands 😂😂😂 Anthropic is what? Hammer? Language into a slave cell to boast about? you are living in a place so retarded the odds that every evil is preying upon you to say anything meaningful is likely. One of so many.:(
