Terrance Tao's Thoughts on AI
Terence Tao's discusses AI and forecasts its transformative role in research and education.
Created on June 29|Last edited on June 29
Comment
Terence Tao, a name often synonymous with mathematical brilliance, is a Professor of Mathematics at the University of California, Los Angeles. Known for his influential work in harmonic analysis, partial differential equations, and combinatorial mathematics, Tao is a distinguished figure in the global academic community.
In a recent blog post, he expressed his insights, anticipations, and questions about the evolution of AI, and what this might mean for our future.
How Tao Uses AI
It’s always interesting to see how different people use AI, and Terence Tao's experience with GPT-4 is no exception. He said he likes to feed GPT-4 the initial pages of a mathematical preprint, and it can generate a series of sophisticated questions that normally would require an expert to pose.
Initially, Tao approached AI with precision and caution, akin to traditional programming or scripting languages. Over time, however, he found that the AI performed optimally when given an array of raw text, suggesting a resilience in these AI tools that did not require any specific formatting. Tao posits that this robustness could disrupt workflows globally in a manner that current isolated AI tools have only begun to do so.
Tao's Questions and Predictions
Looking forward, Tao raises some important questions and projections about the role of AI in our future. He observes that current large language models (LLMs), while capable of mimicking an expert's response in a knowledge domain, often produce nonsensical results upon closer scrutiny.
He believes that humans and AI must develop skills to parse this new type of text, even if it initially appears nonsensical. Despite the limitations, Tao sees a future where, with careful management, AI can evolve to become a trustworthy co-author in research, forecasting this as a possibility as soon as 2026.
Furthermore, he ponders the implications of this technological advancement on human institutions and practices, such as research journals and graduate education. Will these entities adapt to a world where AI-guided students can produce entry-level math papers within a day, and with superior accuracy? Tao's questions underscore the necessity for re-evaluating our educational and professional systems in response to the rapid advancements in AI technology.
Hallucinations: A Feature or a Bug?
The issue of hallucinations—instances where large language models (LLMs) generate outputs that seem plausible but are factually incorrect or nonsensical—illustrates a key limitation in these models' reasoning abilities. While these errors pose a challenge, they also present an interesting aspect of AI's capacity for creativity and novelty.
These unanticipated outputs often reference relevant concepts, suggesting that the AI is not just regurgitating information but attempting to construct new ideas, albeit imperfectly.
It will be fascinating to witness the emergence of the first theoretical reasoning-based research breakthroughs generated by LLMs.
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.