RLHF

English

Noun

RLHF (uncountable)

  1. (machine learning) Initialism of reinforcement learning from human feedback, a method for post-training models.
    Hypernyms: RL, reinforcement learning < DL, deep learning < ML, machine learning < learning
    Coordinate terms: RLAIF, RLVR
    • 2023, Mohak Agarwal, Generative AI for Entrepreneurs in a Hurry‎[1], Notion Press, →ISBN:
      ChatGPT and reinforcement learning with human feedback (RLHF) have revolutionized the AI landscape, providing an accessible and reliable platform for AI-enabled applications.
    • 2024 April 16, Alex Hern, “TechScape: How cheap, outsourced labour in Africa is shaping AI English”, in Katharine Viner, editor, The Guardian‎[2], London: Guardian News & Media, →ISSN, →OCLC, archived from the original on 18 October 2026:
      Turning it into a chatbot requires an extra step, the aforementioned reinforcement learning with human feedback: RLHF. An army of human testers are given access to the raw LLM, and instructed to put it through its paces: asking questions, giving instructions and providing feedback.
    • 2025 May 9, Mike Caulfield, “AI Is Not Your Friend”, in Jeffrey Goldberg, editor, The Atlantic‎[3], Washington, D.C.: The Atlantic Monthly Group, →ISSN, →OCLC, archived from the original on 8 July 2026:
      RLHF now seems more like a process by which machines learn humans, including our weaknesses and how to exploit them. Chatbots tap into our desire to be proved right or to feel special.
    • 2025, Karen Hao, Empire of AI, New York City: Penguin Press, →ISBN:
      At the time, InstructGPT received limited external attention. But within OpenAI, the AI safety researchers had proved their point: RLHF did make large language models significantly more appealing as products.
    • 2025 June 14, Melissa Heikkilä, “AI leaders rein in ‘sycophantic’ chatbots that flatter users”, in FT Weekend (Companies & Markets section), London: The Financial Times Ltd., →ISSN, →OCLC, page 12:
      The “yeasayer effect” arises in AI models trained using reinforcement learning from human feedback (RLHF)—human “data labellers” rate the answer generated by the model as being either acceptable or not.