Abstract
Though large-scale pretrained models have been revolutionary, the time has come for models to start learning like humans. They should be curious about new things and use past knowledge to make educated conjectures. A model should learn untrained tasks on the fly while simultaneously pushing the upper performance bounds of existing knowledge.
Original language | English |
---|---|
Pages | 58-65 |
Number of pages | 8 |
Volume | 58 |
No. | 1 |
Specialist publication | Computer |
DOIs | |
Publication status | Published - 2025 |