An evaluation of three open-source detectors and an exploration of generalisability across datasets and deepfake generation methods, revealing both the adaptability of these models when fine-tuned and their limitations when facing unfamiliar synthetic media.
Timothy Lin
Thoughts from a wandering mind
My main areas of interest includes quantitative research, socio-technological change, open-source software and web technology. I try to combine them in my work in various unique ways.
I am currently a product manager at Resaro and maintainer of a couple of popular open-source projects including Contentlayer, Pliny, and this Tailwind Next.js blog template.
Latest Writing
- Read article
- Read article
A review of key developments in the synthetic data landscape over the past few years, driven by advances in generative AI and falling costs, and a practitioner's perspective on the opportunities and challenges ahead.
- Read article
A guide on how to execute ComfyUI workflows as standalone scripts.
- Read article
Walkthrough on streaming structured objects to create progressively updating interfaces with FastAPI and Next.js.
- Read article
A survey on the different methodologies used to generate structured output from LLMs, from model fine-tuning, to domain specific language, and schema engineering.