PeopleofSuperintelligence

Researcher

Brando Miranda

Brando Miranda is a Stanford PhD researcher whose work connects deep-network theory, language-model evaluation and formal mathematics. Before Stanford, he studied at the University of Illinois Urbana-Champaign and MIT, worked at MIT’s Center for Brains, Minds and Machines (CBMM), and earned an MEng in electrical engineering and computer science under Tomaso Poggio; Stanford awarded him its School of Engineering Fellowship in September 2022. His 2019 research examined how gradient descent can provide implicit complexity control linked to generalization, and with Rylan Schaeffer and Sanmi Koyejo he argued in 2023 that metric choice can create apparent emergent abilities in language models; NeurIPS named the paper an Outstanding Paper. He coauthored work on language-data diversity, Lean 4 autoformalization and mathematical reasoning, including Putnam-AXIOM’s benchmark of 522 competition problems and 100 variations. As of October 2026, he is a PhD student in Stanford’s Trustworthy AI Research group and co-founder and president of Stanford AI for Lean; his September 2026 study reevaluated evidence for open-ended homogeneity in language models.

Now at
Stanford AI for Lean
Badges

Timeline

  1. 2019

    Deep-network generalization

    In 2019, he coauthors “Theory III: Dynamics and Generalization in Deep Networks.” The paper describes gradient descent as providing implicit complexity control linked to generalization.

  2. September 2022

    A Stanford engineering fellowship

    Before Stanford, he studies at the University of Illinois Urbana-Champaign and completes an MEng in electrical engineering and computer science at MIT under Tomaso Poggio, researching deep-learning theory. He works as a research assistant at MIT’s Center for Brains, Minds and Machines (CBMM), and Stanford awards him its School of Engineering Fellowship in September 2022.

  3. 2023

    The NeurIPS mirage paper

    With Rylan Schaeffer and Sanmi Koyejo, he coauthors “Are Emergent Abilities of Large Language Models a Mirage?” The paper shows how metric choice can create apparent emergent abilities; NeurIPS names it an Outstanding Paper in 2023.

  4. 2023–2024

    The diversity coefficient

    In 2023, he coauthors a paper that defines the diversity coefficient to measure variability in natural-language pretraining data. The work appears as a paper at the Data-centric Machine Learning Research Workshop at ICLR 2024.

  5. 2024

    A Lean4 autoformalization benchmark

    In 2024, he coauthors an evaluation benchmark for turning mathematical language into Lean4. The authors test GPT-3.5, GPT-4 and Gemini Pro, and find limits on more complex mathematics.

  6. 2025

    The Putnam-AXIOM benchmark

    In 2025, with Aryan Gulati and five other coauthors, he introduces Putnam-AXIOM, a benchmark of 522 university-level competition problems and 100 generated variations. The paired sets give model researchers a contamination-resilient test bed for mathematical reasoning.

  7. September 2026

    The Artificial Hivemind study

    In September 2026, he coauthors “How Strong Is the Evidence for the Artificial Hivemind? Reevaluating Evidence for the Open-Ended Homogeneity of Language Models.” The study compares same-prompt responses that express different ideas with reported convergence, and finds 20%–32% of those pairs exceed the paper’s 0.8 convergence threshold.

  8. As of October 2026

    Stanford AI for Lean

    As of October 2026, he is a PhD student in Stanford’s Computer Science Department under Sanmi Koyejo in the Trustworthy AI Research group, with interests in theorem proving and meta-learning. He is the co-founder and president of Stanford AI for Lean, a student-led research community focused on AI-assisted theorem proving, Lean 4 and formal verification.

Career

  1. NowStanford AI for LeanCo-founder and President
  2. NowStanford UniversityPhD researcher
  3. Massachusetts Institute of TechnologyResearch Assistant

Education

  1. University of Illinois Urbana-Champaign
  2. Massachusetts Institute of TechnologyMEng in electrical engineering and computer science
  3. Stanford UniversityPhD student

From the biography.

  1. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Stanford University. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.
  2. Google had the most people shaping AI every year from 2014 to 2025Includes Stanford University. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  3. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionIncludes Stanford University. We estimate where the people shaping AI studied and where their careers led. Massachusetts Institute of Technology leads with 467 former students, including 291 recorded doctorates.
  4. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes Stanford University. We estimate where the people shaping AI worked when their most-cited papers appeared: the company share was 31% in 2020–2024 and 8% in 1970–1974, and Google led all organizations in 2020–2024.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Brando Miranda yet. Send us a free-licensed photo.

  1. Theory III: Dynamics and Generalization in Deep Networks arxiv.org. Checked 21 October 2026.
  2. Brando Miranda's Profile | Stanford Profiles profiles.stanford.edu. Checked 21 October 2026.
  3. Brando Miranda cs.stanford.edu. Checked 21 October 2026.
  4. AI’s Ostensible Emergent Abilities Are a Mirage | Stanford HAI hai.stanford.edu. Checked 21 October 2026.
  5. NeurIPS 2023 Awards neurips.cc. Checked 21 October 2026.
  6. Beyond Scale: The Diversity Coefficient as a Data Quality Metric for Variability in Natural Language Data arxiv.org. Checked 21 October 2026.
  7. An Evaluation Benchmark for Autoformalization in Lean4 arxiv.org. Checked 21 October 2026.
  8. Putnam-AXIOM: A Functional & Static Benchmark for Measuring Higher Level Mathematical Reasoning in LLMs proceedings.mlr.press. Checked 21 October 2026.
  9. How Strong Is the Evidence for the Artificial Hivemind? Reevaluating Evidence for the Open-Ended Homogeneity of Language Models arxiv.org. Checked 21 October 2026.
  10. Stanford AI for Lean | Brando Miranda brando90.github.io. Checked 21 October 2026.