Technical articles
Writing
Which Action Deserved the Reward? Exploring Credit Assignment in Reinforcement Learning
An experiment-by-experiment study of whether action-level feedback can improve reinforcement learning for language models.
Read article
Does the LM Head Create a Harmful Gradient Bottleneck? A Causal Test
A response to the LM-head bottleneck claim using backward-only rank controls, repeated-token exposure tests, and small language models.
Read articleMorphology-Aware Reversible Semantic Tokenization and Hierarchical Word Composition for Tamil Language Models
A technical study of reversible Tamil semantic tokenization, hierarchical word composition, and a controlled six-system translation comparison.
Read article
A Quantitative Overview of Modern Tamil Lexicon and Morphology
How fewer than 140,000 root lemmas and inflectional rules generate more than 2.13 billion distinct Tamil written forms.
Read article
Treating Numbers, UUIDs, Dates and Codes as Entities in Language Models: What Worked and What Failed
Experiments with entity registers and typed references for exact values—and why choosing the right entity remained hard.
Read article
Duality, Cause & Effect and Word & Form
An ontological exploration of duality and the fundamental relation between Word and Form.
Read article
சூனியதம்
பௌத்த தத்துவத்தின் மையக் கருத்து
Read article