PeopleofSuperintelligence

Researcher

Boxin Wang

Boxin Wang is an AI researcher whose work spans language-model robustness, trustworthiness, retrieval and multimodal reasoning. He coauthored the NeurIPS 2021 oral benchmark Adversarial GLUE, completed a computer science PhD under Bo Li at the University of Illinois Urbana-Champaign in 2023, and started an NVIDIA internship. His DecodingTrust assessment of GPT models won a NeurIPS 2023 Outstanding Paper award; he also coauthored retrieval-pretrained models such as InstructRetro and multimodal systems such as NVLM and Cosmos-Reason1. As of October 2026, he is a Senior Research Scientist at NVIDIA, where his coauthored Nemotron-Cascade paper was accepted for an oral presentation at NeurIPS 2026.

Now at
NVIDIA

Timeline

  1. 2019–2022

    Adversarial GLUE and robustness

    He coauthors papers on data valuation in 2019 and adversarial text generation in 2020. In 2021, his Adversarial GLUE benchmark tests language-model vulnerabilities to adversarial attacks, and in 2022 he coauthors studies of detoxification, semantic text attacks and certified robustness; NeurIPS lists Adversarial GLUE as an oral.

  2. 2023

    A PhD and NVIDIA internship

    He starts an NVIDIA internship in May 2023. He completes a computer science PhD at the University of Illinois Urbana-Champaign under Bo Li in December 2023, with a dissertation on LLM trustworthiness and adversarial robustness.

  3. 2023

    Retrieval pretraining at EMNLP

    At EMNLP 2023, he coauthors a comprehensive study of RETRO, a scalable retrieval-augmented language model. The study compares RETRO with standard GPT and GPT models that add retrieval at fine-tuning or inference.

  4. 2023

    A NeurIPS paper award

    In 2023, he coauthors DecodingTrust, a comprehensive evaluation of GPT-4 and GPT-3.5 that finds previously unpublished trustworthiness vulnerabilities. NeurIPS names the paper an Outstanding Paper.

  5. 2024

    InstructRetro scales retrieval

    In 2024, he coauthors InstructRetro, which introduces Retro 48B by continuing GPT pretraining with 100 billion additional tokens and retrieving from 1.2 trillion tokens. The paper reports gains on zero-shot tasks after instruction tuning.

  6. 2024

    Multimodal models at NVIDIA

    In 2024, he coauthors NVLM 1.0, NVIDIA's family of multimodal large language models for vision-language tasks. The team reports improved text-only accuracy after multimodal training.

  7. 2025

    Cosmos-Reason1 and physical reasoning

    In 2025, he coauthors Cosmos-Reason1, a multimodal model suite for physical common sense and embodied reasoning. The project builds benchmarks for those capabilities and generates embodied decisions in natural language.

  8. As of October 2026

    A senior researcher at NVIDIA

    He is a Senior Research Scientist at NVIDIA. In September 2026, his coauthored Nemotron-Cascade paper is accepted to NeurIPS 2026 as an oral presentation.

Career

  1. Since 2023NVIDIAIntern

Education

  1. 2023University of Illinois Urbana-ChampaignPhD

From the biography.

  1. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes Boxin Wang. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.
  2. Google has the highest combined h-index among AI organizationsIncludes NVIDIA. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  3. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes NVIDIA. 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.
  4. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes NVIDIA. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.

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 Boxin Wang yet. Send us a free-licensed photo.

  1. Adversarial GLUE: A Multi-Task Benchmark for Robustness Evaluation of Language Models arxiv.org. Checked 20 October 2026.
  2. Boxin Wang - Homepage boxin.wang. Checked 20 October 2026.
  3. Towards trustworthy large language models ideals.illinois.edu. Checked 20 October 2026.
  4. Shall We Pretrain Autoregressive Language Models with Retrieval? A Comprehensive Study aclanthology.org. Checked 20 October 2026.
  5. NeurIPS 2023 Awards neurips.cc. Checked 20 October 2026.
  6. DecodingTrust: A Comprehensive Assessment of Trustworthiness in GPT Models proceedings.neurips.cc. Checked 20 October 2026.
  7. InstructRetro: Instruction Tuning post Retrieval-Augmented Pretraining proceedings.mlr.press. Checked 20 October 2026.
  8. NVLM: Open Frontier-Class Multimodal LLMs research.nvidia.com. Checked 20 October 2026.
  9. Cosmos-Reason1: From Physical Common Sense To Embodied Reasoning arxiv.org. Checked 20 October 2026.