About Me
Hey there! Iām Ben Wyatt. I graduated from the University of Rochester in 2023 with a BS in Physics and BA in Economics.
Academic & Research Experience
During my internship at Los Alamos National Laboratory, I conducted data science work in Python for an experiment at the OMEGAāEP laser facility. The project used very short, intense laser pulses to strike thin film targets, producing fastāmoving proton beams. Our goal was to evaluate how effectively these beams could be used for imaging. Laser-driven proton radiography can penetrate dense materials and capture rapid changes in highāpressure, highātemperature environments, making it valuable for studying fusion experiments and other extreme physics scenarios. Hereās the published paper.
Current Role
Iām currently at EY working as an AI Engineer in New York City. My team is focused on training LLMs for specific internal tasks. Iāve built end-to-end training pipelines for Supervised Finetuning, Continuous Pretraining, and Model Distillation. Most of the grunt work is in the data preparation. Iām also heavily focused on various Synthetic Data Generation techniques, especially multi-page long-context documents.
Iāve also been involved on the infrastructure side, managing the kubernetes deployment on a 16 H200 cluster.
On the application end, Iām a big proponent of the declarative agent framework DSPy.
Some general thoughts about Generative AI
- Enterprise value unlocks require substantial data transformation work. Garbage in, Garbage out!
- The most important organizational knowledge lives tacitly in employeesā brains. For this and other reasons, AI adoption will be slow relative to what some others are saying.
- I love the idea of GraphRAG but havenāt seen an incredible integration yet.
- Small model agentic RL is a very promising innovation and Iām excited to see those tools mature over time.
- Generative AI is already creating immense value, and itās concentrating rapidly in the hands of a small number of highly capable ICs. This is deeply worrying.
Hobbies and Interests
Iāve recently become obsessed with Obsidian for note taking and writing.
I play the guitar and listen to music. Here are some artists Iāve been listening to recently:
Iām really interested in building great knowledge accumulation systems that enable curiosity-driven learning. I like to curate my Substack and RSS feeds. As well as YouTube videos and podcasts. Iāve only recently come around to the utility of Twitter. I see LLMs as a powerful force for this type of curiosity-driven learning and Iām really excited to see what next-generation tools look like.