OpenAI has developed an internal scale to measure the progress of its large language models towards artificial general intelligence (AGI), according to BloomBerg. This five-level classification system was shared with employees during an all-hands meeting and is intended to be released to investors and external parties in the future. The scale serves as a framework to track the advancement of AI capabilities and provides a roadmap for achieving human-level artificial intelligence.

The Five Levels

  1. Level 1: Current conversational AI systems (e.g., ChatGPT)

    • Represents AI that can interact conversationally with humans
    • OpenAI believes its current models are at this level

  2. Level 2: AI capable of solving basic problems at PhD level

    • Called “Reasoners”
    • Can solve problems equivalent to a human with a doctorate-level education, without access to external tools
    • OpenAI is approaching this level

  3. Level 3: AI agents that can perform tasks for users

    • Termed “Agents”
    • Capable of spending several days acting on a user’s behalf
    • Represents a significant leap in AI autonomy

  4. Level 4: AI that can create new innovations

    • Dubbed “Innovators”
    • Demonstrates creative problem-solving and the ability to generate novel ideas

  5. Level 5: AI capable of performing tasks equivalent to organizations of people

    • Referred to as “Organizations”
    • Represents the highest level of AI capability, potentially surpassing human organizational abilities
OpenAI’s AGI Definition
  • Highly autonomous system
  • Surpasses humans in most economically valuable work
  • OpenAI’s mission is to ensure that AGI benefits all of humanity
  • The company acknowledges both the potential benefits and risks associated with AGI development
Key Points
  • OpenAI is currently progressing towards Level 2
  • The company pledges to support other aligned projects if they achieve AGI first
  • Experts’ predictions for AGI achievement vary widely
  • Sam Altman estimated “five years, give or take” in October 2023
  • The classification system is considered a work in progress and may be adjusted based on feedback
Recent Developments
  • Partnership with Los Alamos National Laboratory announced
  • Aim: Investigate how advanced AI models like GPT-4 can contribute to scientific research
  • Goal: Develop a framework for assessing AI capabilities and safety for the US government
  • OpenAI demonstrated a research project involving GPT-4 that shows human-like reasoning skills
Challenges and Concerns
  • AGI realization requires significant computing power investment, potentially costing billions of dollars
  • OpenAI disbanded its safety team in May 2024
  • Concerns raised about prioritizing “flashy products” over safety
  • Debate within the AI community about how to determine the criteria for reaching AGI
Industry Perspective
  • Other companies, like Google DeepMind, have proposed similar frameworks for assessing AI capabilities
  • The AI industry is drawing parallels to the automotive industry’s system for assessing self-driving car automation levels
  • Experts like Matt Wood from Amazon Web Services suggest that AI has not yet reached its exponential growth phase

This expanded presentation provides a more comprehensive overview of OpenAI’s AI power measurement system and its implications for the development of artificial general intelligence.

Via: TheVerge