menu
techminis

A naukri.com initiative

google-web-stories
Home

>

Open Source News

>

AI Society...
source image

Medium

1M

read

306

img
dot

Image Credit: Medium

AI Society: How Networks of Specialized Models Could Shape the Future of Artificial Intelligence

  • As Large Language Models (LLMs) continue to scale, challenges of extreme model scaling are emerging, leading to questions about sustainability.
  • An alternative approach, known as 'AI Society,' suggests using networks of specialized AI models working together like human societies for more efficient AI advancement.
  • Specialized AI models under 100B parameters each offer advantages over larger monolithic models, including better performance in specific domains and more efficient resource utilization.
  • The 'AI Society' approach involves a coordinating system that routes queries to specialized models, exemplified by an implementation called 'Frida,' a home assistant system.
  • Frida consists of orchestrated specialized systems with clear communication protocols, enabling efficient task delegation and reliable outputs.
  • The modular design of Frida allows for extensions with new capabilities without disrupting existing functionality, in contrast to monolithic models requiring expensive retraining.
  • The AI Society model aligns with theories in cognitive science and offers potential implications for achieving more general artificial intelligence through coordination of specialized intelligences.
  • By mimicking the modular architecture of human cognition, AI societies may address safety concerns and be more aligned with human values.
  • This approach democratizes advanced AI capabilities by using smaller open-source models that can run on consumer hardware, making AI more accessible and affordable.
  • Building better societies of AI systems with improved communication and coordination between specialized components is seen as the next frontier in AI development, with potential benefits for all contributors.
  • Shifting the focus from scaling up models to designing efficient networks of specialized AI components may lead to more impactful advancements in artificial intelligence.

Read Full Article

like

18 Likes

For uninterrupted reading, download the app