AI researchers turn to simulated worlds to improve chatbots
Making chatbots better by feeding them more text is becoming increasingly difficult. Researchers are developing systems that learn by moving through s, interacting with objects, and seeing the results of their actions. The aim is to build AI that understands objects and cause and effect, rather than only producing convincing language.
Some environments create scenes in so an can take actions inside them. It remains unproven whether this approach will lead to broad, human-level .
Key points
- Training on more text alone is yielding smaller improvements in chatbots.
- The new approach lets AI learn by acting inside s.
- Researchers want AI to understand objects and cause and effect, not just language.
- An can choose and perform actions within these environments.
- The path from -world skills to human-level is still uncertain.