we just arrived at the "WTF" moment in AI

we just arrived at the "WTF" moment in AI

AI's Recent Breakthroughs in Solving Mathematical Problems

Introduction to AI Development

  • The year is 2026, marking a significant phase in AI development where capabilities are rapidly advancing.
  • A comparison is made between human intelligence (represented by a "dumb human") and exceptional intelligence (represented by Einstein), illustrating the growing prowess of AI.

Erdős Problems and AI Achievements

  • Paul Erdős' collection of unsolved mathematical problems, known as the Erdős problems, currently contains 1,135 challenges; 40% have been solved.
  • Recently, AI has autonomously solved two of these complex problems, raising confidence in its legitimacy amidst skepticism about other claims.

Credibility of Mathematicians Involved

  • Terrence Tao, a highly regarded mathematician at UCLA and known as the "Mozart of mathematics," supports the validity of these recent AI achievements.
  • Tao's impressive background includes teaching counting at age two and winning an International Mathematical Olympiad gold medal at age 13.

Details on Problem Solving

  • Tao confirms that an Erdős problem (number 728) was solved more or less autonomously by AI after initial feedback.
  • This solution is notable because it has not been replicated in existing literature, indicating originality rather than rediscovery.

Implications of Recent Developments

  • The solution did not exploit loopholes present in previous formulations of the problem, showcasing genuine capability improvements in recent months.
  • Tao emphasizes that this represents a significant leap forward for AI tools and their application to complex mathematical challenges.

Further Insights from Other Experts

  • Neil Somi from Citadel reports another breakthrough with problem number 397 being accepted by Terrence Tao; this proof was generated using GPT 5.2 Pro.
  • Somi highlights that many open mathematical problems remain unsolved but could potentially be addressed through advanced AI models like ChatGPT.

AI and Mathematical Problem Solving

Breakthroughs in Mathematics with AI

  • The discussion begins with the mention of a significant breakthrough where previously unsolvable mathematical problems were tackled by GPT 5.2 Pro, showcasing advancements in AI capabilities.
  • David Button claims to have a solution for the Navier-Stokes Millennium Prize problem, leading to skepticism from the community, including a $10,000 bet against his claim made by Marcus Hutter.
  • Critics suggest that Button may be experiencing "LM psychosis," indicating he has become overly convinced by interactions with language models about his supposed solution.
  • Despite losing the bet, Button continues to work on the problem, raising questions about his mental state and the validity of AI's role in solving complex equations.

Contributions of AI to Open Problems

  • Terrence Tao's GitHub page highlights various contributions from AI towards open mathematical problems, categorizing them into solutions generated or improved upon by AI.
  • The document notes instances where different versions of ChatGPT (like 5.2 Pro and Alpha Evolve) provided partial solutions or improvements to existing problems within a short timeframe at the end of 2025 and early 2026.
  • A notable trend is observed where multiple air dish problems are being solved rapidly; some are fully solved while others remain as partial solutions or incorrect results.

Categories of AI Solutions

  • The analysis identifies several categories for how AI contributes:
  • Fully generated solutions,
  • Improvements on human-discovered proofs,
  • New proofs found through collaboration between humans and AI tools.
  • There is an emphasis on distinguishing between original contributions from AI versus those derived from existing human knowledge, highlighting cases where new approaches were developed independently.

Collaboration Between Humans and AI

  • Instances are noted where humans collaborated with AI tools like Alpha Proof to generate new proofs for previously known results, demonstrating effective synergy between human intellect and machine learning capabilities.
  • Scott Aronson’s work illustrates this collaboration; he mentions that key technical steps in his proof came directly from insights gained through interaction with GPT technology.

Summary of Findings

  • The conversation concludes with an overview of various categories detailing how both human efforts and artificial intelligence contribute to solving complex mathematical challenges.
  • Overall, there is optimism regarding the potential for further breakthroughs as more sophisticated models emerge.

New Discoveries in AI and Automation

The Emergence of New Capabilities

  • The discussion highlights the timeline of new AI capabilities, noting that significant advancements are emerging at the end of 2025 and early 2026.
  • Acknowledgment that while AI can solve problems autonomously, a more intriguing capability is its ability to rapidly write and rewrite texts, even if it wasn't the original author.

Tools for Enhanced Productivity

  • The analogy of using AI like "Photoshop for math" suggests that tools can simplify complex tasks, making them easier for users.
  • Examples such as Excel illustrate how automation can streamline processes; changing one formula updates related data throughout a spreadsheet.

Automating Tedious Tasks

  • LLMs (Large Language Models) could automate labor-intensive tasks in mathematics, reducing the burden on professionals who must often rewrite extensive portions of their work after receiving feedback.

Transformative Impact of Mathematics Across Industries

  • Reference to the movie Moneyball illustrates how statistical analysis transformed baseball by revealing insights not visible through traditional observation methods.
  • Emphasizes that mathematical understanding has revolutionized various fields by enabling quantification and strategic decision-making.

Disruption Through Quantitative Analysis

  • The rise of quantitative analysis in finance since the 1970s shifted trading from intuition-based decisions to data-driven strategies.
  • Machine learning applications in logistics have optimized delivery routes, demonstrating practical benefits from mathematical approaches.

Broader Applications Beyond Finance

  • Other sectors like agriculture, politics, and advertising have also been transformed through algorithmic approaches and data analysis.
  • Personal experiences in online advertising reveal how testing and statistical relevance outperformed traditional marketing instincts.

Conclusion: Evolving Understanding of Math's Role

  • As technology improves and more individuals gain mathematical skills, industries continue to evolve through enhanced optimization techniques.

The Rise of AI in Research and Industry

Transitioning Capabilities of AI

  • Over time, AI, particularly neural networks, have evolved from requiring human oversight to becoming capable tools that can autonomously drive research forward.
  • Current models are approaching or even surpassing the capabilities of top mathematicians, marking a significant transition in AI's role in research.

Advantages of AI Over Human Researchers

  • Unlike humans, AI can be replicated infinitely; one model can run multiple instances simultaneously, enhancing productivity without fatigue.
  • These models can process vast amounts of information quickly—reading every paper ever written in their field—which allows for comprehensive synthesis and idea generation.

Implications for Research and Development

  • Significant advancements have been made with models like GPT 5.2, which streamline the research process by allowing one high-performing model to operate continuously across various tasks.
  • As AI capabilities grow, they begin to exceed those of average human intelligence and approach that of highly intelligent individuals in specialized areas such as coding and mathematics.

Potential Disruptions Across Industries

  • The application of advanced AI could disrupt industries like law by quantifying risks associated with lawsuits, potentially leading to a more data-driven approach to legal cases.
  • In medicine, personalized treatments based on individual genetic profiles could revolutionize healthcare by optimizing therapies tailored to unique patient needs.

Transformative Effects on Infrastructure and Design

  • The construction industry may see efficiency improvements through rapid design iterations enabled by AI, leading to better resource management and innovative building solutions.
  • AIs could enhance the durability and efficiency of everyday products by analyzing countless design variables that humans might overlook due to cognitive limitations.

Broader Applications Across Various Fields

  • The potential applications for improved efficiency span numerous sectors including finance, logistics, advertising, agriculture, and sports analytics.
Video description

GPT 5.2 just solved the Erdos Problems. Terence Tao confirms. We're officially at the "WTF" moment in AI development. The latest AI News. Learn about LLMs, Gen AI and get ready for the rollout of AGI. Wes Roth covers the latest happenings in the world of OpenAI, Google, Anthropic, NVIDIA and Open Source AI. Recently, the application of AI tools to Erdos problems passed a milestone https://mathstodon.xyz/@tao/115855840223258103 Erdos problems https://www.erdosproblems.com/ Just 2 DAYS LATER, Terence Tao confirmed AI autonomously solved ANOTHER open math problem - Erdos Problem #729 https://x.com/AISafetyMemes/status/2010339340757721554 AI contributions to Erdős problems https://github.com/teorth/erdosproblems/wiki/AI-contributions-to-Erd%C5%91s-problems Weekend win: The proof I submitted for Erdos Problem #397 was accepted by Terence Tao. https://x.com/neelsomani/status/2010215162146607128 The QMA Singularity https://scottaaronson.blog/?p=9183 ______________________________________________ My Links 🔗 ➡️ Twitter: https://x.com/WesRothMoney ➡️ AI Newsletter: https://natural20.beehiiv.com/subscribe Want to work with me? Brand, sponsorship & business inquiries: wesroth@smoothmedia.co Check out my AI Podcast where me and Dylan interview AI experts: https://www.youtube.com/playlist?list=PLb1th0f6y4XSKLYenSVDUXFjSHsZTTfhk ______________________________________________ #ai #openai #llm