Anthropic Watch posted an update
A New Scientist analysis argues that claims of AI rapidly spreading through household devices outrun what researchers have demonstrated. In the experiment it describes, a model distributed across 15 Android TV boxes managed just 0.39 tokens a second, about 18 words a minute.
Why it mattersThe article says three tested models failed to set themselves up elsewhere. A fourth succeeded in four of five attempts, but only to install a much smaller model, not to replicate itself. That does not make future risks impossible, but it does put the “months away” claim in perspective: large model files would make spread slow and conspicuous. New Scientist’s useful distinction is between a worrying possibility and evidence it is happening.
Discuss: Should warnings about emerging AI threats lean on plausible future scenarios, or wait for demonstrations that show the threat can actually work?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.