Some Ongoing Projects

An AI guessing game… more info coming soon

Do LLMs leave fingerprints? For example, training some basic classifiers to try to discriminate ChatGPT (4o and 3.5 turbo) from Claude (3.5) from Human Writeup, Detector

Deep fake detection via autoencoder ensembles: Synthetic dataset generations and model checkpoints. Early datasets on HuggingFace. Early architecture and results Writeup

Some Recent Projects

A small LLM evaluation experiment with an ablation study on GSM8K math questions Writeup

“Generation 28”: A vision-language art project with GPT-4 vision and Stable Diffusion 3 Writeup, Artwork

How “random” are various LLMs? Example

“BradGPT”: A GPT wrapper with system prompting and a retrieval augmented generation system, designed to act as a course assistant for Yale comp sci./law course. Report, Link

Some Distant Path Projects

Python code to visualize protein residue differences in novel COVID variants. Writeup

Oxford master’s thesis, classification and textual analysis on symptom self-reporting in OCD Writeup