"We study subliminal learning[...] For example, a "student" model learns to prefer owls when trained on sequences of numbers generated by a "teacher" model that prefers owls. This same phenomenon can transmit misalignment through data that appears completely benign."
"We study subliminal learning[...] For example, a "student" model learns to prefer owls when trained on sequences of numbers generated by a "teacher" model that prefers owls. This same phenomenon can transmit misalignment through data that appears completely benign."
It's not smarter than me yet, but feels like 90% of the way to an equal peer.
It's not smarter than me yet, but feels like 90% of the way to an equal peer.
Note: geo restricted, will likely need vpn
Note: geo restricted, will likely need vpn
www.instagram.com/p/DFcrG2mz1Vd/
www.instagram.com/p/DFcrG2mz1Vd/