Aspiring Student-run Research Lab · Est. 2025
Viveka is an aspiring student-led research lab investigating the internal workings of large language models — mechanistic interpretability, reasoning, and the frontiers of AGI evaluation.
Viveka is an aspiring student-run research lab focused on mechanistic interpretability and the evaluation of large language models. We are a group of undergraduate researchers who believe that understanding AI from the inside is one of the most important scientific challenges of our time.
We operate in year-long tenures, each with a focused research agenda, a dedicated team, and a commitment to impactful work. Our research is open, collaborative, and grounded in both theoretical depth and empirical rigor.
We welcome advisors, collaborators, and supporters who share our conviction.
Our inaugural tenure investigates the mechanistic underpinnings of LLM hallucinations, factual recall circuits, and the dynamics of in-context learning. Through probing, activation patching, causal tracing, training toy models in mathematical frameworks, we try to build rigorous explanations of model internals from the ground up.
Explore Viveka 1.0Building on the interpretability foundation of V1.0, our second tenure turns toward frontier AGI evaluation. We interpret and advance performance on ARC-AGI 2, probing how (and whether) models generalise, reason abstractly, and compose solutions to genuinely novel problems.
Explore Viveka 2.0We are grateful for the compute, funding, and mentorship provided by our supporters. If you'd like to support Viveka, reach out.