Mikhail Gromov - 3/4 Generation, Transformation, Transmission, Memorization, Storage and (...)

Mikhail Gromov - 3/4 Generation, Transformation, Transmission, Memorization, Storage and (...)

by Mikhail Gromov

Formal & Physical Sciences AI — Models & Research UYQArtificial intelligence (AI)UYQLNatural language and machine translation
🎙 Mikhail Gromov 👥 79K 📅 April 21, 2026 ⏱ 87 min 👁 6K 📄 lecture 🧭 2026-08-02
Available in: English (current) Français

Keywords

language modelprobability distributionneural networkinformation flowcausality

Summary

In this lecture, Mikhail Gromov explores the mathematical foundations of language and information systems, drawing parallels between biological information processing and artificial neural networks. He begins by describing the probabilistic model of language, where a function assigns probabilities to word sequences, and discusses the challenge of defining such functions in high-dimensional spaces. He then introduces the concept of artificial neural networks as compositions of simple functions, emphasizing the importance of parameter efficiency and generalization. Gromov critiques traditional mathematical linguistics for being irrelevant, and argues that language models capture grammar and semantics in ways that are not fully understood. He connects these ideas to broader themes of prediction, causality, and the orientation of living systems toward the future, contrasting this with physical causality. The lecture touches on information storage, transmission, and reduction in biological systems, from viruses to the brain, and discusses concepts like the central dogma, protein folding, and sparse distributed memory. Gromov also addresses the role of symmetry and modularity in neural circuits, and the stochastic nature of language generation. Throughout, he emphasizes the need for a new mathematical framework to understand these complex systems, and suggests that current models, while successful, remain conceptually opaque.

196 words

Critical Evaluation

Mikhail Gromov’s lecture offers a highly original and thought-provoking perspective on the mathematical underpinnings of language models and information systems. As a Fields medalist, Gromov brings a unique mathematical depth to the discussion, but the lecture is more of a conceptual exploration than a rigorous mathematical treatment. The value of the information lies in its interdisciplinary connections, linking ideas from biology, neuroscience, and computer science to fundamental questions about information and causality. Gromov’s argumentation is solid in its logical flow, but he often relies on intuition and analogies rather than formal proofs, which may leave some points underdeveloped. The scientific rigor is high in terms of the mathematical concepts he invokes, but the speculative nature of some claims, such as the ‘reverse causality’ in biological systems, requires further substantiation. The sources cited are minimal, with only a reference to the Carmin.tv platform, which limits the ability to verify specific claims. The title is incomplete and does not fully capture the content, but the lecture does address the themes of generation, transformation, and transmission of information. Overall, this is a valuable lecture for those interested in the intersection of mathematics, information theory, and cognitive science, but it is not a comprehensive or self-contained introduction to the topic.

206 words

Title / Content Match

The title is incomplete and vague, but the content matches the broader theme of information processing and language models.

Quality & Reliability

8/10

Lecture by a renowned mathematician, presenting a conceptual and mathematical perspective on language models and information systems. The content is speculative and exploratory, but grounded in mathematical reasoning and references to established concepts.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

  • Carmin.tv — Video platform for mathematics and related sciences, hosting this lecture.

Concurring Sources

  • Carmin.tv — Platform hosting the lecture, aligning with the content's academic nature.

Contribution & Novelties

This lecture provides a unique mathematical perspective on language models, connecting them to biological information processing and fundamental questions about causality and information. Gromov’s emphasis on the limitations of traditional mathematical linguistics and the need for new frameworks is a significant contribution.

Pour aller plus loin :

92 words

Radar Profile

The radar profile shows high scores in quantity and quality of information, with a strong technical level, but slightly lower reliability due to the speculative nature of some claims. This suggests a lecture that is rich in ideas but may require critical evaluation.

Reliability 7/10