AGI - The Coming AGI Wars: Players and Positioning
AI Summary
Summary: The Coming War for AI Amongst Giants and Davids
Context: Dr. Aliana Moren, founder of Themosis Incorporated, discusses the competition in the field of artificial general intelligence (AGI) between large companies (Giants) and smaller entities (Davids).
Key Players:
- Large Language Models (LLMs): Not seen as a path to AGI due to their reliance on past data and conditional probabilities.
- OpenAI: 80 billion valuation.
- Google’s Gemini: Investment figures unclear, but Google’s valuation is approximately 30 times that of OpenAI.
- Meta (Facebook): Investing 40 billion in AI, with a market cap of about 1.2 trillion.
- Anthropic and Mistol: Newer companies with significant investments, particularly Anthropics’ $4.5 billion total investment.
- Elon Musk’s X.: Announced a 18 billion valuation.
AGI Approaches:
- Yan LeCun’s Framework: Meta’s Chief Scientist proposes a general architecture for AGI, focusing on reinforcement learning and control theory.
- Carl Friston’s Active Inference: A natural approach to AGI, emphasizing a model that adapts to minimize divergence from the evolving representation of the world.
- Themosis’ Corticons: A new class of neural network that can be used as a model within an active inference framework, potentially powerful for AGI.
Investment and Valuation:
- Meta’s investment in AGI is a fraction of its 40 billion AI budget.
- Versus, the company working with Friston, has a total investment of $21 million, significantly less than other players.
AGI Characteristics:
- Different AGIs will have distinct “feels” and interaction styles, influenced by their underlying philosophies and technologies.
- Sam Altman envisions AGI as either an alter ego or a super-competent colleague.
- The choice of AGI will impact how users interact with and perceive the technology.
Editorial Notes:
- Apologies for omitting perplexity and Google’s Deep Mind from the discussion.
- Clarification on generative models: LeCun suggests AGI may not need generative models, while Friston acknowledges the brain may use a generative approach.
Engagement:
- Themosis offers resources and courses for those interested in AGI and invites engagement through their community and AI salons.
For more detailed information and resources, readers are encouraged to visit the associated blog post and engage with Themosis.