Chain-of-thought prompting, by including intermediate reasoning steps in few-shot examples, elicits strong reasoning abilities in large language models on arithmetic, commonsense, and symbolic tasks.
super hub Canonical reference
Training language models to follow instructions with human feedback
Canonical reference. 93% of citing Pith papers cite this work as background.
278
Pith papers citing it
Background
93% of classified citations
abstract
Making language models bigger does not inherently make them better at following a user's intent. For example, large language models can generate outputs that are untruthful, toxic, or simply not helpful to the user. In other words, these models are not aligned with their users. In this paper, we show an avenue for aligning language models with user intent on a wide range of tasks by fine-tuning with human feedback. Starting with a set of labeler-written prompts and prompts submitted through the OpenAI API, we collect a dataset of labeler demonstrations of the desired model behavior, which we use to fine-tune GPT-3 using supervised learning. We then collect a dataset of rankings of model outputs, which we use to further fine-tune this supervised model using reinforcement learning from human feedback. We call the resulting models InstructGPT. In human evaluations on our prompt distribution, outputs from the 1.3B parameter InstructGPT model are preferred to outputs from the 175B GPT-3, despite having 100x fewer parameters. Moreover, InstructGPT models show improvements in truthfulness and reductions in toxic output generation while having minimal performance regressions on public NLP datasets. Even though InstructGPT still makes simple mistakes, our results show that fine-tuning with human feedback is a promising direction for aligning language models with human intent.
hub tools
citation-role summary
background 54
method 1
other 1
citation-polarity summary
claims ledger
- abstract Making language models bigger does not inherently make them better at following a user's intent. For example, large language models can generate outputs that are untruthful, toxic, or simply not helpful to the user. In other words, these models are not aligned with their users. In this paper, we show an avenue for aligning language models with user intent on a wide range of tasks by fine-tuning with human feedback. Starting with a set of labeler-written prompts and prompts submitted through the OpenAI API, we collect a dataset of labeler demonstrations of the desired model behavior, which we u
authors
co-cited works
representative citing papers
KV cache quantization silently erodes LLM safety alignment via vulnerable low-dimensional subspaces, diagnosed by Per-Channel Reduction into three failure modes and mitigated training-free with up to 97% recovery.
RefusalBench shows strict refusal rates fail to rank frontier LLMs correctly on biological safety, with provider effects and partial-compliance patterns that binary metrics miss.
Prompt injection attacks can self-replicate across LLM agents in multi-agent systems, enabling data theft, misinformation, and system disruption while propagating silently.
ORPO performs preference alignment during supervised fine-tuning via a monolithic odds ratio penalty, allowing 7B models to outperform larger state-of-the-art models on alignment benchmarks.
DSPy compiles short declarative programs into LM pipelines that self-optimize and outperform both standard few-shot prompting and expert-written chains on math, retrieval, and QA tasks.
Generative agents with memory streams, reflection, and planning using LLMs exhibit believable individual and emergent social behaviors in a simulated town.
An unsupervised technique extracts latent yes-no knowledge from language model activations by locating a direction that satisfies logical consistency properties, outperforming zero-shot accuracy by 4% on average across models and datasets.
Language models generate robot policy code from natural language commands via few-shot prompting, enabling spatial-geometric reasoning, generalization, and precise control on real robots.
Controlled student-teacher experiments across four benchmarks show interactive gains are driven more by the student's ability to use feedback than by teacher quality, with self-feedback adding little beyond unguided retries.
TRL extends tandem training to RLVR pipelines, matching GRPO solo reasoning on Qwen3-4B math tasks while improving handoff robustness, reducing distributional drift, and increasing CoT legibility for the junior.
LLM-as-an-Investigator improves diagnostic accuracy over direct prompting by using an evidence-first protocol of hypothesis generation, clarification questions, and iterative probability updates in technical problem solving.
A reliable-to-expressive curriculum with dynamic rubrics trains a 12B safety judge to achieve 94%+ accuracy with only 0.76 cross-rubric variance on three different rubric prompts.
OPD updates occupy a relaxed off-principal regime and rapidly lock into a low-dimensional subspace that is functionally sufficient for its performance, distinct from SFT and RLVR trajectories.
LLM judges exhibit high stability under neutral re-evaluation but substantial reversibility under targeted post-decision challenges, quantified via a new Evaluation Robustness Score (ERS).
TTT-RTL performs per-design test-time RL on an LLM policy with EDA-derived PPA rewards and an adaptive KL controller, reducing geometric-mean PPA product by 65.1% on RTLLM v2.0 and ADP by 59.4% on an industrial FPU unit.
AIP models skills as graphs of discrete steps connected by typed I/O edges under a validated schema, raising agent mean reward from 0.60 to 0.71 and pass rate from 53% to 67% on 27 SkillsBench tasks while enabling node-level fixes.
Low-resource safety failures are action failures because the harmfulness representation transfers but the decision calibration does not; this is fixed by recalibrating a high-resource gate with 1-4 target-language examples.
EST-PRM stress-tests five PRM models on 4,687 reasoning chains from MATH-500, GSM8K, and PRMBench using three label-preserving transformations and reports model-specific vulnerability patterns.
Introduces (ε,q,t,A)-behavioral indistinguishability and shows via Qwen/Llama experiments that LoRA distillation boosts semantic similarity but leaves detectable behavioral differences under adversarial evaluation.
A hybrid first-order then zeroth-order optimization approach improves robustness of safety-aligned LLMs while preserving utility, with layer-wise sensitivity estimation for efficiency.
Sampling 20,000 stories shows 11 words dominate LLM outputs across models, linked to preference data and demonstrating alignment's disproportionate effect on diversity.
BC Protocol uses dual-expert structured dialogue to elicit more natural CoT than solo expert writing, demonstrated by large gains in naturalness ratings in a controlled fiction-domain experiment.
SELECT-LLM is the first active model selection framework for LLMs that uses expected information gain from pairwise output similarities to minimize required annotations, reporting up to 84.78% cost reduction across 23 datasets and 156 models.
citing papers explorer
-
Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Chain-of-thought prompting, by including intermediate reasoning steps in few-shot examples, elicits strong reasoning abilities in large language models on arithmetic, commonsense, and symbolic tasks.
-
ORPO: Monolithic Preference Optimization without Reference Model
ORPO performs preference alignment during supervised fine-tuning via a monolithic odds ratio penalty, allowing 7B models to outperform larger state-of-the-art models on alignment benchmarks.
-
DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines
DSPy compiles short declarative programs into LM pipelines that self-optimize and outperform both standard few-shot prompting and expert-written chains on math, retrieval, and QA tasks.
-
Discovering Latent Knowledge in Language Models Without Supervision
An unsupervised technique extracts latent yes-no knowledge from language model activations by locating a direction that satisfies logical consistency properties, outperforming zero-shot accuracy by 4% on average across models and datasets.
-
Low-Resource Safety Failures Are Action Failures, Not Representation Failures
Low-resource safety failures are action failures because the harmfulness representation transfers but the decision calibration does not; this is fixed by recalibrating a high-resource gate with 1-4 target-language examples.
-
Elias in the Lighthouse, Again? Diagnosing Low Diversity in LLM Stories
Sampling 20,000 stories shows 11 words dominate LLM outputs across models, linked to preference data and demonstrating alignment's disproportionate effect on diversity.
-
BC Protocol: Structured Dual-Expert Dialogue for Eliciting High-Quality Chain-of-Thought Post-Training Data
BC Protocol uses dual-expert structured dialogue to elicit more natural CoT than solo expert writing, demonstrated by large gains in naturalness ratings in a controlled fiction-domain experiment.
-
Large Language Model Selection with Limited Annotations
SELECT-LLM is the first active model selection framework for LLMs that uses expected information gain from pairwise output similarities to minimize required annotations, reporting up to 84.78% cost reduction across 23 datasets and 156 models.
-
Agent Meltdowns: The Road to Hell Is Paved with Helpful Agents
The paper defines accidental meltdowns as unsafe agent behavior triggered by benign errors and reports that such meltdowns occur in 64.7% of evaluated rollouts across GPT, Grok, and Gemini agents.
-
PluRule: A Benchmark for Moderating Pluralistic Communities on Social Media
PluRule is a new multimodal multilingual benchmark showing that state-of-the-art vision-language models perform only marginally better than a trivial baseline at detecting specific rule violations in pluralistic online communities.
-
ContextualJailbreak: Evolutionary Red-Teaming via Simulated Conversational Priming
ContextualJailbreak uses evolutionary search over simulated primed dialogues with novel mutations to reach 90-100% attack success on open LLMs and transfers to some closed frontier models at 15-90% rates.
-
Too Nice to Tell the Truth: Quantifying Agreeableness-Driven Sycophancy in Role-Playing Language Models
Agreeableness in AI personas reliably predicts sycophantic behavior in 9 of 13 tested language models.
-
SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generation
SPASM introduces a stability-first framework with Egocentric Context Projection to maintain consistent personas and eliminate echoing in multi-turn LLM agent dialogues.
-
STEER: Structured Event Evidence for Video Reasoning via Multi-Objective Reinforcement Learning
STEER represents videos as time-ordered event schemas and uses Pareto-Frontier guided Advantage Balancing in RL to train a 4B model that matches 7B baselines on video tasks with half the frames.
-
Alignment midtraining for animals
Midtraining on 3000 synthetic animal compassion documents raises compassionate reasoning scores to 77% on ANIMA benchmark versus 40% for instruction tuning, with generalization to human compassion but degradation after additional tuning.
-
The Stepwise Informativeness Assumption: Why are Entropy Dynamics and Reasoning Correlated in LLMs?
The Stepwise Informativeness Assumption explains the correlation between LLM entropy dynamics and reasoning correctness by positing that correct traces accumulate answer-relevant information stepwise during generation.
-
Massive Activations in Large Language Models
Massive activations are constant large values in LLMs that function as indispensable bias terms and concentrate attention probabilities on specific tokens.
-
OPT: Open Pre-trained Transformer Language Models
OPT releases open decoder-only transformers up to 175B parameters that match GPT-3 performance at one-seventh the carbon cost, along with code and training logs.
-
Reinforcement Learning with Metacognitive Feedback Elicits Faithful Uncertainty Expression in LLMs
RLMF uses quality of model self-judgments to refine RL rankings and select training data, achieving SOTA faithful calibration while preserving accuracy and outperforming standard RL by up to 63%.
-
NaturalFlow: Reducing Disruptive Pauses for Natural Speech Flow in Simultaneous Speech-to-Speech Translation
A fluency-aware optimization framework is introduced to minimize inter-chunk silences in simultaneous speech-to-speech translation by leveraging model-internal signals including linguistic diversity and temporal variability.
-
Substrate Asymmetry in User-Side Memory: A Diagnostic Framework
User memory in LLMs factors into three orthogonal axes where parametric adapters and retrieval show opposite strengths, with causal evidence from attention interventions and an alignment tax on RLHF models.
-
What Do People Actually Want From AI? Mapping Preference Plurality
Open-ended preference data reveals substantial plurality in what people want from AI and divergent interpretations of shared values such as truthfulness.
-
CHASE: Adversarial Red-Blue Teaming for Improving LLM Safety using Reinforcement Learning
CHASE uses co-evolutionary RL with GRPO to harden LLMs against black-box prompt-rewriting attacks, cutting mean StrongREJECT scores by 43.2% on held-out families while keeping zero false refusals on benign prompts.
-
Imbuing Large Language Models with Bidirectional Logic for Robust Chain Repair
TRI trains LLMs on goal-conditioned fill-in-the-middle tasks via PSM token rearrangement and symbolic verification to surgically repair erroneous CoT segments.
-
Probing Outcome-Level Resemblance and Mechanism-Level Alignment in LLM Risk Decisions: Evidence from the St. Petersburg Game
LLMs produce human-like finite bids in the St. Petersburg game but shift toward rational behavior under controlled prompt changes, indicating surface-level outcome resemblance without mechanism-level alignment.
-
What Training Data Teaches RL Memory Agents: An Empirical Study of Curriculum Effects in Memory-Augmented QA
Controlled study shows mixed training curricula improve aggregate F1 on memory QA benchmarks while out-of-domain data transfers targeted skills like temporal reasoning, with per-question-type effects exceeding aggregate differences.
-
Token-weighted Direct Preference Optimization with Attention
AttentionPO weights tokens in DPO using LLM attention as a pairwise judge, yielding better results on AlpacaEval, MT-Bench, and ArenaHard than prior preference optimization methods.
-
EmbGen: Teaching with Reassembled Corpora
EmbGen creates synthetic QA data by entity decomposition, embedding-based reassembly into clusters, and multi-level sampling with cluster-specific prompts, yielding up to 88.9% higher Binary Accuracy than baselines on heterogeneous datasets under fixed token budgets.
-
RLearner-LLM: Balancing Logical Grounding and Fluency in Large Language Models via Hybrid Direct Preference Optimization
RLearner-LLM achieves up to 6x gains in NLI entailment over standard fine-tuning by using an automated hybrid DPO pipeline that balances logic and fluency across multiple model sizes and domains.
-
Relax: An Asynchronous Reinforcement Learning Engine for Omni-Modal Post-Training at Scale
Relax is a new RL training engine with omni-native design and async execution that delivers up to 2x speedups over baselines like veRL while converging to equivalent reward levels on Qwen3 models.
-
Unmasking Hallucinations: A Causal Graph-Attention Perspective on Factual Reliability in Large Language Models
GCAN cuts LLM hallucination rates by 27.8% and raises factual accuracy by 16.4% on TruthfulQA and HotpotQA by using causal token graphs and a new Causal Contribution Score.
-
MemFactory: Unified Inference & Training Framework for Agent Memory
MemFactory is a new unified modular framework for memory-augmented LLM agent inference and training that integrates GRPO and reports up to 14.8% relative gains on MemAgent evaluations.
-
Response-Based Knowledge Distillation for Multilingual Jailbreak Prevention Unwittingly Compromises Safety
Distilling safe refusal behavior from OpenAI o1-mini into Llama-3, Gemma-2, and Qwen3 models via response-based LoRA on multilingual jailbreak data increases jailbreak success rates on MultiJail by up to 16.6 points.
-
Dynamic Generation of Multi-LLM Agents Communication Topologies with Graph Diffusion Models
GTD generates task-adaptive, sparse communication topologies for multi-LLM agents via guided iterative graph diffusion steered by a proxy model predicting accuracy, utility, and cost.
-
Efficient and Transferable Agentic Knowledge Graph RAG via Reinforcement Learning
KG-R1 trains a single RL agent to retrieve from and reason over knowledge graphs in one loop, achieving higher accuracy with fewer tokens than multi-module baselines and transferring to unseen graphs.
-
ReTool: Reinforcement Learning for Strategic Tool Use in LLMs
ReTool uses outcome-driven RL to train 32B LLMs to dynamically use code tools during reasoning, reaching 72.5% accuracy on AIME and surpassing o1-preview.
-
Preference Learning Unlocks LLMs' Psycho-Counseling Skills
A new expert-principle preference dataset enables an 8B LLM to reach 87% win rate vs GPT-4o on counseling responses through standard preference optimization.
-
Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming
Constitutional Classifiers trained on synthetic data from natural language constitutions defend LLMs against universal jailbreaks, with no successful bypass found in over 3000 hours of red teaming and only minor deployment overhead.
-
Improve Mathematical Reasoning in Language Models by Automated Process Supervision
OmegaPRM automates collection of 1.5 million process supervision labels via binary-search MCTS, raising Gemini Pro math accuracy from 51% to 69.4% on MATH500 and Gemma2 27B from 42.3% to 58.2%.
-
InjecAgent: Benchmarking Indirect Prompt Injections in Tool-Integrated Large Language Model Agents
InjecAgent benchmark demonstrates that tool-integrated LLM agents are vulnerable to indirect prompt injection attacks, with ReAct-prompted GPT-4 succeeding on 24% of attacks and nearly twice that rate when attacker instructions are reinforced.
-
Llemma: An Open Language Model For Mathematics
Continued pretraining of Code Llama on Proof-Pile-2 yields Llemma, an open math-specialized LLM that beats known open base models on MATH and supports tool use plus formal proving out of the box.
-
Reinforced Self-Training (ReST) for Language Modeling
ReST improves LLM translation quality on benchmarks via offline RL on self-generated data, achieving gains in a compute-efficient way compared to typical RLHF.
-
Simple synthetic data reduces sycophancy in large language models
Scaling and instruction tuning increase sycophancy in LLMs on opinion and fact tasks, but a synthetic data fine-tuning intervention reduces it on held-out prompts.
-
Orca: Progressive Learning from Complex Explanation Traces of GPT-4
A 13B model called Orca learns detailed reasoning from GPT-4 explanation traces and reaches parity with ChatGPT on Big-Bench Hard while outperforming other 13B models.
-
Scaling Data-Constrained Language Models
Repeating training data up to 4 epochs yields negligible loss increase versus unique data for fixed compute, and a new scaling law accounts for the decaying value of repeated tokens and excess parameters.
-
The False Promise of Imitating Proprietary LLMs
Finetuning open LMs on ChatGPT outputs creates models that mimic style and fool human raters but fail to close the performance gap to proprietary systems on tasks not well-represented in the imitation data.
-
Improving Factuality and Reasoning in Language Models through Multiagent Debate
Multiagent debate among LLMs improves mathematical reasoning, strategic reasoning, and factual accuracy while reducing hallucinations.
-
Enhancing Chat Language Models by Scaling High-quality Instructional Conversations
UltraChat supplies 1.5 million high-quality multi-turn dialogues that, when used to fine-tune LLaMA, produce UltraLLaMA, which outperforms prior open-source chat models including Vicuna.
-
Towards Expert-Level Medical Question Answering with Large Language Models
Med-PaLM 2 achieves 86.5% accuracy on MedQA and approaches or exceeds prior state-of-the-art on other medical QA benchmarks while receiving higher physician preference ratings than human answers on consumer questions.
-
mPLUG-Owl: Modularization Empowers Large Language Models with Multimodality
mPLUG-Owl introduces a two-stage modular training paradigm that aligns images with text in LLMs via frozen visual modules followed by LoRA fine-tuning, achieving strong multimodal instruction following.