Naveed Ahmad

Nous Research Proposes Lighthouse Attention: A Training-Only Selection-Based Hierarchical Attention That Delivers 1.4–1.7× Pretraining Speedup at Long Context

Training large language models on long sequences has a well-known problem: attention is expensive. The scaled dot-product attention (SDPA) at the core of every transformer scales quadratically Θ(N²) in both compute and memory with sequence length N. FlashAttention addressed this through IO-aware tiling that avoids materializing the full N×N attention matrix in high-bandwidth memory, reducing…

Read More

Trump Adds Coinbase and Bitcoin Stocks to Portfolio

Broader tech allocations in the same filing show a parallel bet on AI and infrastructure alongside crypto-linked assets. A federal financial disclosure filed by Donald Trump on May 14 shows his portfolio purchased shares of MARA Holdings, Coinbase, and Strategy between January and March 2026. Out of more than 3,600 transactions listed across 113 pages,…

Read More