NVIDIA and OpenAI Are Scaling an AI That Cannot Subtract

2026-06-09 · Source: AI Advances - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

The article asserts that contemporary AI, despite substantial investments, including "\$1 trillion" by companies like NVIDIA and OpenAI, operates on "broken arithmetic" that hinders its ability to perform basic operations such as subtraction, leading to errors like "2 + 2 = 2". The author contends that the foundational mathematics, including linear algebra, optimization, and tensors, currently underpinning AI development may be fundamentally flawed. This suggests a critical need to re-evaluate the core mathematical principles guiding AI, potentially rendering existing paradigms obsolete. The text briefly alludes to an alternative geometric approach, emphasizing the importance of metrics, connections, and holonomy for maintaining meaning, contrasting it with the limitations of "flat embeddings".

Key takeaway

For AI Scientists and Machine Learning Engineers focused on foundational research, you should critically re-evaluate the mathematical underpinnings of current AI models. Your understanding of linear algebra and optimization might be insufficient for future advancements if the "broken arithmetic" claim holds true. Consider exploring alternative mathematical frameworks, particularly those involving geometry, to address fundamental limitations in AI's ability to handle basic logical operations. This could prevent investing further in potentially flawed paradigms.

Key insights

Current AI's mathematical foundations are flawed, leading to basic arithmetic failures despite massive investment.

Principles

• Current AI relies on "broken arithmetic."
• Standard linear algebra may be insufficient.
• Geometry offers an alternative to "flat embeddings."

Topics

• AI Foundations
• Mathematical Limitations
• Geometric Deep Learning
• NVIDIA
• OpenAI
• Arithmetic Errors

Best for: AI Scientist, Machine Learning Engineer, AI Student

Related on AIssential

• The Roadmap of Mathematics for Machine Learning and AI — Mastering core mathematics is crucial for advancing beyond baseline machine learning performance.
• Are NVIDIA, Anthropic, and Google Hiding the Real AI? — Current AI's fundamental geometric problem cannot be solved by scaling flat transformers; Geometric AI offers a more efficient solution.
• The Mathematics of AI Winters: The mathematical Taxonomy of Paradigm Fragility in AI Winter — AI winters were fundamentally influenced by mathematically precise bottlenecks, not just engineering or commercial failures.
• Shape, Symmetries, and Structure: The Changing Role of Mathematics in Machine Learning Research — Mathematics' role in ML is shifting from theoretical guarantees to post-hoc explanations and high-level architectural guidance, leveraging advanced pure math.
• An OpenAI model has disproved a central conjecture in discrete geometry — An OpenAI model disproved a long-standing mathematical conjecture using algebraic number theory, marking a significant AI reasoning milestone.
• OpenAI’s Biggest Blunder: Scaling the Math That Makes ChatGPT Bluff — More computation does not mean AI stops hallucinating.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Advances - Medium.