AI
Anthropic self-replication study splits the technology press
Anthropic's research on whether small models can set themselves up on cheap, internet-connected hardware has divided coverage. The experiment showed a nine-billion-parameter model establishing inference on eight of nine hardware types and a larger open-weight model running across fifteen generic Android streaming boxes, with researchers configuring every deployment themselves. Coverage split between framing the result as evidence that self-replication is months away and accusing the lab of conjuring a threat it concedes it has not observed in the wild.