Terrence Tao's ChatGPT Conversation about the Jacobian Conjecture Counterexample
Terrence Tao, a Fields Medalist, engaged ChatGPT in a highly technical mathematical dialogue, leading to a sophisticated counterexample for the long-standing Jacobian Conjecture. This fascinating interaction showcases AI's potential as a profound research assistant for even world-leading experts, pushing the boundaries of what these models can achieve. It has sparked fervent discussion on Hacker News about the nature of AI intelligence, its role in human discovery, and the future of specialized knowledge.
The Lowdown
Terrence Tao, a renowned mathematician, used ChatGPT to explore the Jacobian Conjecture, a notoriously difficult open problem in mathematics for decades. The conversation, which is shared publicly, demonstrates a sophisticated interaction where Tao, with his deep domain expertise, guided the AI through complex algebraic geometry and polynomial theory.
- Tao's prompts were highly specific and technical, pushing the AI to consider various aspects of the problem, including searching for simpler sub-results and generalizing ideas.
- The AI, in turn, provided structured responses, performed calculations, ingested PDFs, wrote code, and even 'thought' and 'pushed back' with caveats, ultimately helping to construct a specific polynomial counterexample in three dimensions.
- The collaboration resulted in a tangible output: a formal proof of a counterexample to a specific formulation of the Jacobian Conjecture, formalized in Lean (a proof assistant). This particular counterexample leverages a complex polynomial structure rather than a brute-force approach.
- The conversation highlights how an expert can leverage an LLM not just for information retrieval, but for an iterative, exploratory research process, effectively turning the AI into a highly capable, albeit guided, mathematical 'colleague.'
This interaction serves as a compelling demonstration of AI's burgeoning role as an 'intelligence amplifier,' enabling top-tier researchers to accelerate discovery by offloading computationally intensive or pattern-matching tasks, thereby allowing humans to focus on higher-level intuition and conceptual guidance.
The Gossip
AI as an Expert's Amplifier
Commenters were deeply impressed by how Terrence Tao, a supergenius, skillfully leveraged ChatGPT. Many observed that the AI acted as an amplifier for his existing expertise, not an independent discoverer. Tao's precise, jargon-filled prompts and ability to steer the conversation were key, showing that mastery of the problem domain is crucial for extracting maximum utility from LLMs, essentially turning AI into a powerful research copilot.
The 'Keep Going' Prompt & LLM Peculiarities
A recurring point of fascination was the effectiveness of simple prompts like 'keep going' or 'what's next' in pushing LLMs beyond their initial responses or perceived limitations. This led to discussions about how LLMs give up prematurely or require specific directives to continue iterating, even for complex problems. Some found humor in these almost childlike prompts, while others suggested sophisticated ways to embed such instructions for better automation.
Mathematical Language and Labyrinthine Logic
A significant portion of the discussion revolved around the inherent density and perceived impenetrability of advanced mathematical notation and nomenclature. Many non-mathematicians confessed feeling utterly lost, comparing it to 'uncommented software' or a 'linguistic swamp.' Mathematicians countered that this abstraction is necessary for reasoning about complex problems, but acknowledged the difficulty for outsiders and even specialists outside their niche, suggesting AI might help bridge these gaps or translate concepts.
Debating the Nature of AI Intelligence
The discussion invariably spiraled into philosophical debates about whether LLMs truly possess 'intelligence' or if they are 'just' next-token predictors. Some argued that the apparent intelligence is undeniable, regardless of the underlying mechanism, especially when witnessing such complex problem-solving. Others insisted that while impressive, it's merely a sophisticated tool lacking genuine consciousness or understanding, leading to arguments about how intelligence should be defined and measured, and AI's potential to 'dull' human intellect over time.
The Jacobian Conjecture Context & Implications
Commenters provided context on the Jacobian Conjecture, noting its notoriety for false proofs and its 'cognitohazard' status for LLMs, given their training data often predates such discoveries. There was discussion about the significance of this 'discovery'—whether it was a revolutionary breakthrough or a crucial 'brick' in the larger edifice of mathematical knowledge, and how AI's current 'knowledge cutoff' influences its ability to tackle unsolved problems.