Post
In 1957, the linguist J. R. Firth observed that ‘you shall know a word by the company it keeps’. The principle that words sharing contexts share meaning underlies generative AI, from early Latent Semantic Analysis to today's largest Transformers. This post traces that lineage with three interactive LSA-to-PCA visualisations in R, built from Reuters newswire, State of the Union addresses and IMDB reviews. They show where simple co-occurrence models succeed and where they fail, and the post explains how scale and the Transformer turned a modest insight into the technology behind ChatGPT. It then examines why LLMs, optimised for fluency, cannot be rid of hallucination entirely, and argues that careful prompt engineering is the best tool we have for steering a fundamentally heuristic machine.
Post
A transcription pipeline, written with the help of GitHub Copilot, runs OpenAI's Whisper on local machines or high-performance computing clusters. Hundreds of hours of sensitive interview audio can be transcribed this way without sending any data to the cloud, and each hour of recording now takes minutes to process.
Post
A production-ready local transcription workflow leveraging OpenAI's Whisper models that addresses the limitations of cloud-based solutions through complete data sovereignty, unlimited scale, reproducible processing and advanced quality control, while maintaining GDPR compliance.
Software Pipeline
A secure, local transcription workflow using OpenAI's Whisper, with batch processing on high-performance computing (HPC) clusters and built-in quality control. Designed for academic and enterprise use, it avoids the privacy and reproducibility limitations of cloud-based services.