Spotlight: Feb 7, 2023
Large language models like GPT-3 can learn new tasks without updating their parameters — despite not being trained to perform those tasks. A new study “could explain almost all of the learning phenomena that we have seen with these large models,” says Ekin Akyürek.