

Is GPT-3 Dissapointing, Killer Robots (?), the AI Hall of Shame
Jun 10, 2021
Dive into the latest AI controversies! Discover why GPT-3 is being called a disappointment compared to its predecessor. Explore King County's historic ban on facial recognition technology. The rise of self-driving trucks completing long hauls faster than humans raises eyebrows. Plus, hear chilling discussions about the implications of autonomous robots in warfare. And don't miss the exploration of AI failures in the newly created Hall of Shame, urging tech accountability and ethical considerations.
AI Snips
Chapters
Transcript
Episode notes
GPT-3's Disappointing Paper
- GPT-3 paper demonstrates few-shot learning abilities but lacks conceptual transformations.
- It pushes the field forward by showing qualitative improvements in language models.
True Few-Shot Learning
- True few-shot learning doesn't use validation data for model selection, unlike previous approaches.
- This reveals that prior few-shot learning results might be overestimated due to reliance on validation sets.
AI-Piloted Planes
- Merlin Labs, backed by Google Ventures, develops autonomous flight systems for airplanes.
- The company has conducted hundreds of unmanned test flights, aiming to become the leading autonomy platform for aircraft.