Inference‑Time Steering for Cross‑Lingual Factual Consistency in LLMs
Comparing prompting, activation steering, and preference tuning for more consistent factual answers across languages.
Selected work
Research
& writing.
Language models, security, and the systems behind useful AI.
Publications
Controlling Cross‑Lingual Answer Distributions in Language Models: Enabling Transfer of Factual Preferences
Steering answer distributions across languages and testing how those interventions transfer to cultural settings.
Read PDF Related to the study above.
Tamper‑Evident Pairing
An examination of secure wireless pairing, its verification, and the conditions behind a man‑in‑the‑middle vulnerability.
Thesis
Applications of Large Language Models in Cybersecurity
Evaluating language models for classification, data augmentation, and synthetic data in spam detection and policy compliance.
Read PDF 82 pages
Writing
The Future of Agentic AI
A case for smaller, specialised models and intelligence built across an agentic system.