AIFEATURE

How RLHF Trains Language Models to Align With Human Preferences

林紀旭 James LinEditor-in-Chief
Published · Updated
RLHF trains a reward model on human preference rankings, then uses reinforcement learning to fine-tune language models toward that reward signal. OpenAI's InstructGPT research demonstrates the three-step method in practice: human labelers preferred outputs from the 1.3-billion-parameter InstructGPT over the 175-billion-parameter GPT-3, a model with 100 times more parameters.

What Is RLHF, and Why Align Language Models to Human Values?

RLHF trains a reward model with direct human feedback, then uses it to optimize an AI agent's performance through reinforcement learningCITE:E7. IBM defines the technique this way, describing it as a machine learning method in which a reward model is trained on human feedback and subsequently used to improve an AI agent's behavior through reinforcement learningCITE:E7. Hugging Face states that RLHF has enabled language models to begin aligning a model trained on a general corpus of text data to complex human valuesCITE:E6.

What Is the First Step, and What Role Does Supervised Learning Play?

The first step fine-tunes GPT-3 on labeler-written demonstrations of desired behaviorCITE:E1. In the InstructGPT paper published March 4, 2022, OpenAI describes collecting a dataset of labeler demonstrations of the desired model behavior and using it to fine-tune GPT-3 with supervised learningCITE:E1. This supervised model becomes the starting point for the next two steps.

What Is the Second Step — How Is a Reward Model Trained From Human Ranking Data?

The second step trains a reward model on human rankings of model outputsCITE:E2. OpenAI collects a dataset of rankings of model outputs and uses it to further fine-tune the supervised model with reinforcement learning from human feedbackCITE:E2. Hugging Face explains that the underlying goal of this reward model is to take in a sequence of text and return a scalar reward that numerically represents the human preferenceCITE:E4.

What Is the Third Step — How Does PPO Apply in Reinforcement Learning Fine-Tuning?

The third step applies Proximal Policy Optimization, or PPO, to fine-tune the language model against the reward signalCITE:E5. Hugging Face describes this step as fine-tuning some or all of the parameters of a copy of the initial language model using a policy-gradient reinforcement learning algorithm, PPOCITE:E5. This step builds directly on the ranking data collected in step two, since OpenAI uses that dataset to further fine-tune the supervised model through RLHFCITE:E2.

What Did InstructGPT Show Compared to GPT-3?

Human labelers preferred outputs from the 1.3-billion-parameter InstructGPT model over the 175-billion-parameter GPT-3CITE:E3.

ModelParametersLabeler Preference
InstructGPT1.3BPreferred
GPT-3175BLess preferred
Parameter gapInstructGPT has 100x fewer parameters than GPT-3

OpenAI reports that despite having 100 times fewer parameters, the 1.3B InstructGPT model's outputs were preferred to those of the 175B GPT-3CITE:E3.

What This Means

The three-step process OpenAI describes in the InstructGPT paper — supervised fine-tuning on demonstrationsCITE:E1, reward model training on ranked outputsCITE:E2CITE:E4, and PPO-based reinforcement learningCITE:E5CITE:E2 — matches the general RLHF framework defined by IBMCITE:E7 and the alignment goal described by Hugging FaceCITE:E6. The InstructGPT result, where a 1.3B-parameter model outperformed a 175B-parameter model on human preference despite 100 times fewer parametersCITE:E3, is the concrete outcome tied to that three-step pipeline in this evidence set.

📊 Evidence

FAQ

What Is RLHF, and Why Align Language Models to Human Values?

RLHF trains a reward model with direct human feedback, then uses it to optimize an AI agent's performance through reinforcement learningCITE:E7.

What Is the First Step, and What Role Does Supervised Learning Play?

The first step fine-tunes GPT-3 on labeler-written demonstrations of desired behaviorCITE:E1.

What Is the Second Step — How Is a Reward Model Trained From Human Ranking Data?

The second step trains a reward model on human rankings of model outputsCITE:E2.

What Is the Third Step — How Does PPO Apply in Reinforcement Learning Fine-Tuning?

The third step applies Proximal Policy Optimization, or PPO, to fine-tune the language model against the reward signalCITE:E5.

📎 Sources

  1. arxiv.org
  2. huggingface.co
  3. ibm.com

Related data

Author's Take林紀旭 James Lin

The InstructGPT result is the strongest evidence here for why RLHF matters as a training architecture, not just a fine-tuning add-on: a 1.3-billion-parameter model beat a 175-billion-parameter model on human preference despite carrying 100 times fewer parameters. That gap did not come from scale — it came from the three-step pipeline itself: supervised fine-tuning on demonstrations, a reward model trained on ranked outputs, and PPO optimization against that reward signal. For anyone weighing investment in an RLHF pipeline against simply scaling a base model, the metric worth tracking is preference-win rate per parameter, not raw parameter count or benchmark scores. Whether the same 100x pattern holds outside the specific InstructGPT setup described here is a question this evidence set does not answer.

林紀旭 James LinEditor-in-Chief

Related

BRIEF

GreenTrans Unveils GT5X, GT3X Quadruped Robots, Targets 100% Taiwan-Made Content by 2027

GreenTrans (綠捷), the robotics subsidiary of China Motor (中華車), unveiled quadruped robots GT5X and GT3X at SEMICON Taiwan 2026, targeting 100% Taiwan-made content by 2027. The robots combine an in-house-designed control unit and battery management system, NVIDIA's Jetson Orin and Isaac Lab platforms, and a new LFP battery developed with Formosa Smart Energy (台塑新智能), while GreenTrans's inspection robots are already deployed in semiconductor fabs.

EffectStory 編輯部 ·
BRIEF

Nvidia Confirms $12.93 Billion Acquisition of Hugging Face

Nvidia confirmed on September 3, 2026 that it agreed to buy Hugging Face for $12.93 billion, exactly $12,930,300,000, gaining the open-source AI hosting platform used by over 18 million developers. CEO Jensen Huang pledged the platform will stay open, with no Nvidia compute required to build on or deploy through it.

EffectStory 編輯部 ·
BRIEF

NVIDIA to Subscribe US$3.5 Billion of MediaTek's Record US$3.9 Billion Convertible Bond

NVIDIA will subscribe US$3.5 billion of MediaTek's US$3.9 billion offshore convertible bond offering, the largest such issuance in Taiwan's capital market history, deepening cooperation in AI infrastructure, edge AI computing, and automotive platforms while marking NVIDIA's first major investment in a Taiwanese company.

EffectStory 編輯部 ·