William Zhou

Experience

• 

  UCLA

  Aug 2022 - Jun 2024

  ML researcher at Professor Bolei Zhou's lab for Human-Friendly AI Interactions. a. Wrote five tutorials for a mathematical introduction to Human-Friendly Reinforcement Learning series.b. Building a library of offline Reinforcement Learning (RL) benchmarks for the Metadrive environment, a general-purpose self-driving environment for evaluating models on predictability and safety. Developed machine learning (ML) models to analyze the relationship between ovarian cancer cell features and cancer progression to produce lab-testable correlations. • Advised peers at UCLA on career fairs and professional networking opportunities.• Reviewed dozens of resumes and provided feedback.

  • Researcher

    Jun 2023 - Jun 2024

  • Researcher

    Aug 2022 - Jul 2023

  • Career Peer Intern

    Aug 2022 - Mar 2023
• 

  Global Challenges Project

  Nov 2022 - Nov 2022

  AI Safety and X-Risk at UC Berkeley

  Invited to GCP's AI Safety and X-risk Workshop at Berkeley's group of 20 from across America.- Discussed ideas from bio-risk and eliciting latent knowledge (ELK) to the anthropic principle.

• 

  The Centre for Effective Altruism

  Feb 2023 - Feb 2023

  Effective Altruism Global 2023

  Attended the Effective Altruism Global 2023 (EAG) conference in Oakland on cost-effective interventions, existential risk, and AI safety and interpretability research.

• 

  Consensus

  Jun 2023 - Oct 2023

  AI Research Intern

  AI Intern at Consensus NLP, a company dedicated to revolutionizing scientific search using natural language processing (NLP) technologies.• Boosted Consensus Search’s query speed by over 80% without losing performance via knowledge distillation of machine learning (ML) models in the search pipeline.• Built the prompting behind Consensus's ChatGPT plugin used in 5M+ conversations.• Improved Consensus Meter’s absolute accuracy by 5% by re-training old models with emerging techniques (Function-Matching Distillation and Stochastic Weight Averaging).• Fixed the Consensus Meter’s tendency to agree with common misconceptions and improved the ability of the pipeline to respond to unusually formatted queries. Show less

• 

  Consensus

  Jun 2024 - Sept 2024

  AI Research Intern

  AI Intern at Consensus NLP, a company dedicated to revolutionizing scientific search using natural language processing (NLP) technologies.• Built GPT-4 scaffolding to generate, benchmark, and QA fine-tuning training examples.• Optimized the Consensus Yes/No meter with query-aware models in the search pipeline.

• 

  Timaeus

  Jun 2024 - Sept 2024

  ML Research Engineering Intern

  MLE Intern at Timaeus, an independent AI Safety research organization on Singular Learning Theory.• Accelerated research by building internal tooling for automating the research experiment pipeline.• Built a visualization toolkit for Singular Learning Theory experiments.• Added support for multiprocessing, multi-GPU, and TPU setups.