I'm Isaiah Freeman, an AI researcher working on mechanistic interpretability and representation engineering. I hold a B.S. and M.S. in Computer Science from Virginia State University, and I run independent research under Freethought Labs.
I want to know what language models represent internally, whether those representations are shared across models, and how we can read and steer them to make models more transparent and more trustworthy.
- Cross-model representation transfer — do different LLMs encode concepts like truth in compatible ways, and can a direction learned in one model be carried into another?
- Probing hidden states — reading a model's decisions (e.g. which tool it's about to call) from its activations before it says anything
- Mixture-of-Experts routing — what expert routing traces reveal about what a model is processing
- Representation-level control — steering and inference-time unlearning by intervening directly on internal features
- Shared Truth: Emergent Truth Properties in Large Language Models via Heterogeneous Injection-based Transfer — NeurIPS 2026 (poster), first author. Transfers truth representations across 24 model pairs and finds that a closed-form Procrustes alignment matches trained adapters on most pairs.
- Pre-Verbal Tool Selection (working paper, Freethought Labs) — reading a model's tool choice from a single mid-stack hidden state.
- RepProbe (open source) — tooling for probe-based hidden-state analysis.
- OSMoSIS (M.S. thesis) — codebook-driven LLM labeling distilled into a student-response sufficiency classifier for educational feedback.
- Preparing the Shared Truth camera-ready and presenting in Atlanta
- Applying to PhD programs in CS (interpretability / representation learning)
- 👯 Open to collaborating on interpretability and representation engineering research
