Intelligence

Investigated what intelligence actually is—not behavior, but memory-prediction. Compared biological and artificial approaches to determine if transformers understand or just pattern-match at scale.

Apollo AI (crop disease detection)

A mini-project for crop disease detection for AI500 hackathon: ResNet-18 model, claimed 95% accuracy, and Telegram deployment for Uzbek farmers.

Continual Learning (in LLMs) - [in progress]

Examined whether LoRA and full fine-tuning produce equivalent solutions despite structural differences. Used recent NeurIPS research showing that parameter-efficient methods achieve similar performance through fundamentally different learned representations. Analyzed implications for continual learning claims and the architectural constraints that determine what LLMs can 'remember' versus relearn.

Creative Writing Handbook - Atlas: Getting the Words right

Analyzed more than 20 writing-related books to gather together the best creative writing concepts and organize them in a proper order. Understood that simple sentences are harder to write than complex ones (they expose thinking rather than obscure it).

Knowledge Fragmentation (silk road narratives)

Explored how historical knowledge emerges from multiple, competing narratives rather than singular accounts. Used Foucault's framework to analyze the Silk Roads as a case study in fragmented knowledge systems and decentralized information networks.