PeopleofSuperintelligence

Researcher

Danny Hernandez

Danny Hernandez is an AI researcher whose work measured training compute and algorithmic efficiency and examined model alignment, self-knowledge, and reasoning. At Twitch, he shaped personalized discovery and built a forecasting culture before consulting for Open Philanthropy and joining OpenAI’s Foresight team. His research explored language assistants and methods for making them helpful and harmless. In August 2023, he was focused on AI alignment research.

Now at
OpenAI
Works
18
Citations
2,517
h-index
16

Publication numbers from OpenAlex, as of 19 October 2026.

Timeline

  1. 2015

    Personalizing Twitch discovery

    As a Twitch data scientist, he presents the recently watched feature as his biggest win of 2015; the experiment increases viewing time by 1.1% for its treatment group.

  2. 2016

    Teaching Twitch to forecast

    In 2016, he builds a culture of predictions at Twitch and uses reviewed forecasts to track participants’ perceived improvement.

  3. By 2020

    Measuring AI progress

    By 2020, he works on AI progress measurement on OpenAI’s Foresight team. With Dario Amodei, he develops AI and Compute, his first OpenAI project, which reports 300,000-fold growth in compute used for major AI training runs between 2012 and 2018.

  4. 2020

    Quantifying algorithmic efficiency

    In 2020, he and Tom Brown report that the compute required to train an ImageNet classifier to AlexNet-level performance falls 44-fold from 2012 to 2019.

  5. 2021–2022

    Training helpful assistants

    In 2021, he coauthors research on language assistants aligned with human values. In 2022, he coauthors work applying reinforcement learning from human feedback and Constitutional AI to train helpful, harmless assistants.

  6. 2022

    Predicting model knowledge

    In 2022, he coauthors a study showing that language models can estimate whether their own claims are valid and predict whether they know an answer.

  7. July 2023–August 2023

    Studying faithful reasoning

    In July 2023, he coauthors studies of whether chain-of-thought explanations reflect a model’s actual reasoning and whether question decomposition can improve faithfulness. By August 2023, he focuses on AI alignment research.

Career

  1. 2022–2023Machine Intelligence Research Institute
  2. 2020OpenAI
  3. From 2015TwitchData scientist
  4. Open PhilanthropyConsultant

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models2022 · 569 citationsWith Abhinav Rastogi, Abubakar Abid, Andrea Madotto, Ashish Sabharwal, Chris Callison-Burch and Christopher D. Manning
  2. Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback2022 · 417 citationsWith Catherine Olsson, Chris Olah, Dawn Drain, Liane Lovitt and Zac Hatfield-Dodds
  3. AI-to-AI Code Reviews of GitHub Pull Requests2022 · 319 citationsWith Cameron McKinnon, Carol Chen, Catherine Olsson, Chris Olah, Dawn Drain and Ethan Perez
  4. Predictability and Surprise in Large Generative Models2022 · ACM Conference on Fairness, Accountability, and Transparency (FAccT) · 206 citationsWith Catherine Olsson, Chris Olah, Dawn Drain, Liane Lovitt and Zac Hatfield-Dodds
  5. Discovering Language Model Behaviors with Model-Written Evaluations2023 · 188 citationsWith Cameron McKinnon, Catherine Olsson, Chris Olah, Dawn Drain, Ethan Perez and Liane Lovitt

From OpenAlex, as of 19 October 2026.

  1. The median h-index is 53 at universities and 19 at frontier AI labsIncludes Danny Hernandez. The median h-index is 53 at universities and 19 at frontier AI labs. We compare publishing careers, citations and moves to estimate how universities and frontier labs contribute to AI research.
  2. 13 of 79 people shaping AI with a former OpenAI role now work at AnthropicIncludes OpenAI. Our estimates of where OpenAI recruits came from and where people with former OpenAI roles work now.
  3. Google alumni founded or led 87 other organizations, the most among the labs we compareIncludes OpenAI. We trace 212 organizations founded or led by AI lab alumni, and 223 career moves between the labs.
  4. 44% of AI career moves go from one university to anotherIncludes OpenAI. We estimate how 14,272 AI career moves link universities, companies, labs and government research centers, with net flows and the most common paths.

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 Danny Hernandez yet. Send us a free-licensed photo.

  1. Recently Watched: A Data Story blog.twitch.tv. Checked 19 October 2026.
  2. Why I’m Teaching Twitch to Predict the Future blog.twitch.tv. Checked 19 October 2026.
  3. Danny Hernandez on forecasting and the drivers of AI progress 80000hours.org. Checked 19 October 2026.
  4. Measuring the Algorithmic Efficiency of Neural Networks arxiv.org. Checked 19 October 2026.
  5. A General Language Assistant as a Laboratory for Alignment arxiv.org. Checked 19 October 2026.
  6. Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback anthropic.com. Checked 19 October 2026.
  7. Constitutional AI: Harmlessness from AI Feedback arxiv.org. Checked 19 October 2026.
  8. Language Models (Mostly) Know What They Know arxiv.org. Checked 19 October 2026.
  9. Measuring Faithfulness in Chain-of-Thought Reasoning arxiv.org. Checked 19 October 2026.
  10. Question Decomposition Improves the Faithfulness of Model-Generated Reasoning arxiv.org. Checked 19 October 2026.
  11. How quickly is AI advancing? And should you be working in the field? (with Danny Hernandez) podcast.clearerthinking.org. Checked 19 October 2026.