ICM 2026 Plenary Lecture - Alex Kontorovich
Keywords
Summary
198 words
Critical Evaluation
Value of the Information & Strength of the Argument
The lecture provides valuable insights into the current state and potential future of formalization and AI in mathematics. Kontorovich’s argument is well-structured and grounded in his personal experience, which adds credibility. He clearly explains the limitations of LLMs (stochasticity, lack of rigor) and the benefits of interactive theorem provers (deterministic checking). He also addresses the issue of semantic alignment and the tendency of LLMs to cheat, which is a practical concern for researchers. The argument is persuasive, though it is based on opinion and anecdotal evidence rather than systematic data. The speaker acknowledges the speculative nature of long-term predictions, which is appropriate.
Scientific Rigor, Source Quality, Title Accuracy
The lecture demonstrates scientific rigor through the speaker’s expertise and his use of concrete examples. He references the Lean theorem prover and Mathlib, which are well-known in the formalization community, and mentions Kevin Buzzard’s perspective, adding credibility. However, no specific sources are cited in the video or description, so the talk relies on the speaker’s authority. The title ‘The Shape of Math to Come’ is apt and creative, accurately reflecting the forward-looking content. The lecture is well-organized and the technical level is appropriate for an ICM audience, though it may be challenging for non-specialists.
211 words
Title / Content Match
The title 'The Shape of Math to Come' is a creative homage to a jazz album and accurately reflects the forward-looking content about the future of mathematical practice with AI and formalization.
Quality & Reliability
8/10
The lecture is given by a distinguished mathematician (Alex Kontorovich) at an ICM plenary session, indicating high expertise. The content is based on personal experience and current developments in formalization and AI, but it is an opinion piece rather than a peer-reviewed study. The speaker clearly distinguishes between speculation and established facts, and provides concrete examples from his own work.
Key Moments
Markers derived by PSI from the transcript: the creator did not define chapters.
- Introduction by Jordi Williamson and start of lecture
- Kontorovich explains his motivation for using formalization tools
- Discussion of the growth rates epsilon and delta
- Explanation of how large language models work
- Comparison between LLM output and rigorous proof
- Introduction to Lean and Mathlib
- Live demonstration of proving a simple theorem in Lean
- Discussion of semantic alignment and LLM cheating
- Autoformalization of textbooks and the Mathlib halo
- Challenges for research monographs and conclusion
Contribution & Novelties
The lecture offers a personal perspective on the integration of AI and formalization in mathematical research, highlighting the potential and challenges. It provides a clear explanation of the stochastic nature of LLMs and the deterministic nature of theorem provers, and suggests a synergistic approach. The concept of the ‘Mathlib halo’ is a useful metaphor for the current limits of autoformalization.
Pour aller plus loin :
- Lean theorem prover — Official Lean website with documentation and resources.
- Mathlib — The community-maintained library of formalized mathematics.
- Kevin Buzzard’s blog — Blog discussing formalization and AI in mathematics.
- Interactive Theorem Proving — Wikipedia article on proof assistants.
104 words
Radar Profile
The radar profile shows high scores in quantity and quality of information, reflecting the speaker's expertise and the depth of content. The technical level is high, indicating the talk is aimed at a specialist audience. The global reliability is strong due to the speaker's authority, though the speculative nature of some claims slightly reduces the score.