Last Week in AI #115 - GPT4, Bard, AGI, Alpaca, Anthropic, Midjourney V5
8 snips
Mar 26, 2023 In a deep dive into the latest AI advancements, GPT-4 emerges as a powerhouse, showcasing exciting new capabilities, including its role as a virtual volunteer. Google unveils Bard, raising eyebrows about its competitiveness. The ethical realm comes into play as discussions evolve around AI safety and transparency while addressing troubling layoffs in responsible AI teams. Midjourney V5 impresses with its art generation skills, prompting an exploration of creativity's future in the age of machines. Plus, groundbreaking methods for enhancing database efficiency shake up the tech landscape!
AI Snips
Chapters
Transcript
Episode notes
GPT-4's Capabilities
- OpenAI's GPT-4 surpasses ChatGPT with image and text inputs, achieving 90th percentile on LSAT.
- This exemplifies unpredictable capability emergence in AI scaling, solving tasks previously deemed impossible.
GPT-4's Opacity
- GPT-4's paper lacks training details due to competition, unlike GPT-3's transparency.
- This raises concerns about the balance between open research and competitive advantage in AI development.
Multimodality and Transfer Learning
- GPT-4 maintains performance with multimodal input, avoiding negative transfer seen in earlier models.
- Scaling appears to overcome this limitation, raising questions about future AI capabilities.