Jeremy Baron leads machine-learning engineering at Levy Street. Jeremy works on model evaluation, research systems and the parts of client projects where off-the-shelf approaches are not good enough.
Background
Jeremy has more than ten years of experience in AI algorithms and previously worked as an engineer at a quantitative hedge fund. Jeremy wrote TorchWindow, a Python library that displays PyTorch tensors straight from GPU memory.
Research
Jeremy is the author of Tiny Shakespeare CLM, a character-level language model written entirely as Python and improved by an AI agent editing its code, which cut training-set bits per character from 4.77 to 2.31 across 463 commits. The approach, which the project calls Code Policy Optimization, is described in Code Policy Models. Jeremy is a joint first author of GTM-Bench.
Work
- Tiny Shakespeare CLMA character-level language model written entirely as Python rules, improved by an AI agent over 463 commits.
- Code Policy ModelsA research direction: improve a policy written as readable code with LLM-proposed edits instead of gradient descent.
- GTM-BenchA 72-task benchmark for AI agents doing go-to-market research, published by Blackpearl Group and co-authored by Levy Street team members.
- Open sourceWorld of ClaudeCraft, SPAWN D, Tiny Shakespeare CLM and TorchWindow: what each is, its licence and its status.