Why, in Spite of their Trillions, Current AI Leaders will not win the AGI Race

2025-05-01 · Source: Peter’s Substack · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Entrepreneurship & Start-ups · Depth: Intermediate, short

Summary

Leading AI companies are fundamentally committed to Big Data architectures like Large Language Models (LLMs) and backpropagation-based technologies, evidenced by their massive datacenter investments. However, these approaches are inherently limited in adaptability and dependability, and cannot achieve Artificial General Intelligence (AGI). The core argument is that AGI demands continuous, real-time, incremental learning, which backpropagation-based systems cannot provide as they are "frozen" after bulk training. Furthermore, these brute-force methods are costly and inefficient compared to a first principles approach based on human cognition. The article contends that current leaders cannot change direction due to mono-culture myopia, incentive structures tied to existing LLM investments, and a lack of a first principles theory of AGI, making them unable to find viable alternative architectures.

Key takeaway

For investors evaluating AI startups, recognize that current LLM-centric leaders face architectural and incentive barriers to AGI. Your due diligence should prioritize startups demonstrating a viable, defensible theory and roadmap to AGI, a working prototype, and a team with deep, relevant expertise in non-backpropagation architectures. This shift could lead to a paradigm change, where a lean, adaptive AGI upstart rapidly gains dominant market share over encumbered giants.

Key insights

Current Big Data AI architectures cannot achieve AGI due to their inability to learn incrementally.

Principles

• AGI requires continuous, real-time, incremental learning.
• Backpropagation systems cannot learn incrementally.
• First principles approach to AGI is based on human cognition.

In practice

• Focus on AGI architectures enabling continuous learning.
• Prioritize cost-efficient, local-device capable AI solutions.

Topics

• AGI Race
• Big Data AI
• Large Language Models
• Backpropagation
• Incremental Learning

Best for: Entrepreneur, Investor, Director of AI/ML

Related on AIssential

• The next step towards AGI — AI progress is accelerating due to diverse innovations, shifting the focus from raw scale to sample efficiency and data compression.
• Why Building Walls Won't Win the AI Race — Global AI leadership hinges on rapid innovation and R&D investment, not on restrictive knowledge barriers.
• The AI Scaling Problem — Current AI benchmarks misrepresent intelligence by prioritizing knowledge application over continuous knowledge acquisition.
• Deepseek reportedly prioritizes AGI research over quick profits despite billions in funding — Deepseek prioritizes AGI and open-source development over short-term profits, even with significant state-backed funding.
• The AI market is unlikely to settle into a simple winner-take-all outcome. Instead, AI will diffuse through the economy as a set of competing ecosystems, each winning different segments... — AI adoption hinges on "relative net value," not just raw model power, leading to a fragmented market.
• The Real Reason AGI Will Never Happen... Hear Me Out — The pursuit of AGI via current scaling methods faces insurmountable physical constraints in power and heat dissipation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Peter’s Substack.