Mathy AI Substack
Subscribe
Sign in
Home
Notes
Archive
About
The Complexity Dynamics of Grokking
Mike’s Daily Paper Review – April 30, 2025
May 1, 2025
•
Mike Erlihson, Mathy AI
7
2
1
Latest
Top
Discussions
The Geometry of Probabilities in LLMs
Mike’s Daily Paper Review: 22.02.26, Review 582, THE INFORMATION GEOMETRY OF SOFTMAX: PROBING AND STEERING
10 hrs ago
•
Mike Erlihson, Mathy AI
2
1
Merging via Lie Algebra: Preserving Manifold Geometry in Deep Model Merging
Mike’s Daily Paper Review: 20.02.26, Review 581; Orthogonal Model Merging
Feb 20
•
Mike Erlihson, Mathy AI
6
1
LLMs as Operators on Measures: A Clean Mathematical Pipeline for Tokenization, Training, Attention, and Decoding
Mike’s daily paper review: 18.02.26LARGE LANGUAGE MODELS: A MATHEMATICAL FORMULATION
Feb 18
•
Mike Erlihson, Mathy AI
4
The “Aha” That Wasn’t: When Language Models Only Pretend to Think
Omri and Mike’s Daily Paper, 16.02.26, Review 579; The Illusion of Insight in Reasoning Models
Feb 16
•
Mike Erlihson, Mathy AI
and
Omri Sapir
4
Bees, Flowers, Deep in the Web
Shlomi and Mike’s Daily Paper, 13.02.26; Interpretability of Graph Neural Networks to Assess Effects of Global Change Drivers on Ecological Networks
Feb 13
•
Mike Erlihson, Mathy AI
and
Shlomi Bereznik
4
1
The Thermodynamics of Transformer Errors: A New Physics of Scale
Mike’s daily paper review: 11.02.26 , A MODEL OF ERRORS IN TRANSFORMERS, Review 577
Feb 11
•
Mike Erlihson, Mathy AI
2
From World Cup Statistics to Business Data: AI21’s New RAG
Shmulik and Mike’s Daily Paper Review 09.02.26, Structured RAG for Answering Aggregative Questions
Feb 9
•
Mike Erlihson, Mathy AI
and
Shmulik Cohen
9
2
See all
Mathy AI Substack
Deep learning paper reviews, blogs about machine learning techniques, LLM discussions
Subscribe
Recommendations
View all 11
AI by Hand ✍️
Prof. Tom Yeh
Interconnects AI
Nathan Lambert
Vital Statistics
Kareem Carr, PhD
The Gradient
The Gradient
AI Tidbits
Sahar Mor
Mathy AI Substack
Subscribe
About
Archive
Recommendations
Sitemap
This site requires JavaScript to run correctly. Please
turn on JavaScript
or unblock scripts