Eric Drexler revisits his 2019 report, «Reframing Superintelligence,» in light of recent advances in Large Language Models (LLMs). He argues that the report’s abstract model of AI development still holds true, even with the unexpected divergence of LLM-based technologies. The core idea is to broaden the understanding of superintelligence beyond just superintelligent agents, including AI systems providing comprehensive services. This reframing emphasizes different risks and opportunities, suggesting a more accessible, transparent, and manageable form of superintelligence through open agencies rather than unitary agents.
What is the main idea of «Reframing Superintelligence»?
Expanding the ontology of superintelligence; includes AI services; emphasizes structures and relationships.
This perspective promotes the Comprehensive AI Services (CAIS) model, viewing general intelligence as a property of flexible service systems.
What is the Comprehensive AI Services (CAIS) model?
General intelligence as flexible services; task-focused agents; expands toward superintelligent performance.
What is a key difference between AI services and AI agents?
AI services are task-focused; AI agents are unitary; AI services offer safety affordances.
Artículo Original: https://www.alignmentforum.org/posts/LxNwBNxXktvzAko65/reframing-superintelligence-llms-4-years
Advices:
- Focus on developing AI services rather than unitary AI agents to potentially create safer and more manageable AI systems.
- Prioritize research into AI R&D automation to improve AI development processes and refocus safety concerns on expanding safe AI functionality.
- Explore the use of techniques like fine-tuning and reinforcement learning to ensure task focus and bounded behavior in AI systems with broad world knowledge.
Deja una respuesta