arc-blog

a simple whitespace theme for academics

Can Mamba Learn, Unlearn, and Retain Noise?

Extending the SLM Noise Study to State Space Models — Mamba 1.4B vs Transformers across 4 noise types

18 min read · February 12, 2026

2026 · ai, mamba, ssm, transformers, noise, unlearning · ai
Multi-Hop Reasoning in Transformers

A Journey From Confidence to Confusion to Clarity: How RoPE Position Embeddings Enable Length Generalization in Transformer Reasoning

20 min read · January 17, 2026

2026 · ai, llm, transformers, rope, reasoning, mechanistic-interpretability, position-embeddings · ai
ReLU vs GELU

Impact of Activation Curvature on Transformer Stability: GELU vs. ReLU

5 min read · January 07, 2026

2026 · ai, llm, gpt, gelu, relu · ai
hello blog

initial blog post

1 min read · November 07, 2025

2025 · personal