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.
259
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.
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.
Distribution-Aware Reward optimizes LLM regression by treating rollouts as empirical predictive distributions and rewarding marginal improvements in CRPS quality rather than point accuracy alone.
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.
citing papers explorer
-
Alignment Collapse Under KV Cache Quantization: Diagnosis and Mitigation
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.
-
On the Geometry of On-Policy Distillation
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.
-
Alpha-RTL: Test-Time Training for RTL Hardware Optimization
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.
-
EST-PRM: Stress-Testing Process Reward Models Before They Become Load-Bearing
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.
-
Bounded Behavioral Indistinguishability for Black-Box LLM Distillation
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.
-
Distribution-Aware Reward: Reinforcement Learning over Predictive Distributions for LLM Regression
Distribution-Aware Reward optimizes LLM regression by treating rollouts as empirical predictive distributions and rewarding marginal improvements in CRPS quality rather than point accuracy alone.
-
Learning, Fast and Slow: Towards LLMs That Adapt Continually
Fast-Slow Training uses context optimization as fast weights alongside parameter updates as slow weights to achieve up to 3x better sample efficiency, higher performance, and less catastrophic forgetting than standard RL in continual LLM learning.
-
Leveraging Pretrained Language Models as Energy Functions for Glauber Dynamics Text Diffusion
Pretrained language models are used as energy functions for Glauber dynamics in discrete text diffusion, improving generation quality over prior diffusion LMs and matching autoregressive models on benchmarks and reasoning tasks.
-
Discrete Tilt Matching
Discrete Tilt Matching recasts dLLM fine-tuning as state-level matching of tilted local unmasking posteriors, producing a stable weighted cross-entropy loss that improves Sudoku and Countdown performance when applied to LLaDA-8B-Instruct.
-
S-GRPO: Unified Post-Training for Large Vision-Language Models
S-GRPO unifies SFT and RL for LVLMs via conditional ground-truth injection that supplies a maximal-reward anchor when group exploration fails completely.
-
Reinforcement Learning via Value Gradient Flow
VGF solves behavior-regularized RL by transporting particles from a reference distribution to the value-induced optimal policy via discrete value-guided gradient flow.
-
CapTrack: Multifaceted Evaluation of Forgetting in LLM Post-Training
CapTrack shows post-training causes drift beyond facts, with instruction fine-tuning producing stronger behavioral changes than preference optimization across model families.
-
Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach
A recurrent-depth architecture enables language models to improve reasoning performance by iterating computation in latent space, achieving gains equivalent to much larger models on benchmarks.
-
Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads
Medusa augments LLMs with multiple decoding heads and tree-based attention to predict and verify several tokens in parallel, yielding 2.2-3.6x inference speedup via two fine-tuning regimes.
-
Let's Verify Step by Step
Process supervision significantly outperforms outcome supervision for training models on the MATH dataset, achieving 78% accuracy on a representative test subset with active learning and a released 800k step-label dataset.
-
Retroactive Advantage Correction: Closed-Form V-Trace Bias Correction for Delay-Aware RLHF
RAC is a closed-form bias correction for delayed rewards in RLHF that is unbiased under full mass reinjection of the delay kernel and reduces to V-trace with no delay.
-
RASFT: Rollout-Adaptive Supervised Fine-Tuning for Reasoning
RASFT is an adaptive SFT method that strengthens or relaxes expert imitation per problem based on on-policy rollout solvability and adds clipped reference-policy ratio to limit drift, reporting better results than standard SFT and RL on math and code benchmarks.
-
SALT: When More Rollouts Don't Help in Group-Based Policy Optimization and How to Make Them Matter
SALT is a subspace-adaptive plug-in for GRPO that decomposes group-relative coefficients into shared and residual channels using mini-batch Gram geometry and amplifies residuals to mitigate signed cancellation in RLVR.
-
Towards Efficient and Evidence-grounded Mobility Prediction with LLM-Driven Agent
AgentMob is a training-free LLM-driven agent that formulates mobility prediction as adaptive evidence-controlled decision making and outperforms other training-free LLM methods on three datasets.
-
Annealed Softmax Greedy in Many-Armed Bayesian Bandits
Annealed softmax greedy achieves Õ(m + T/m) Bayes regret (Õ(√T) at m=Θ(√T)) in many-armed Bayesian Bernoulli bandits under linear upper-tail prior condition, matching empirical-mean greedy.
-
Understanding Goal Generalisation in Sequential Reinforcement Learning
Empirical analysis of over 100 sequential RL training pipelines across 250+ OOD environments finds salient features drive generalization and early goals persist, with latent policy gradients simulating latent variable evolution to predict OOD behavior from training history.
-
When Are Teacher Tokens Reliable? Position-Weighted On-Policy Self-Distillation for Reasoning
Position-Weighted On-Policy Self-Distillation (PW-OPSD) weights later tokens more heavily after a diagnostic shows position predicts teacher reliability better than entropy, yielding +1.0 and +1.1 Avg@12 gains on AIME 2024/2025.
-
PRISM: Preference-Aware Influence Function Based Data Selection Method for Efficient Fine-Tuning
PRISM weights target examples by model preference to build an improved direction for influence-based data selection in LLM fine-tuning.
-
TimeSRL: Generalizable Time-Series Behavioral Modeling via Semantic RL-Tuned LLMs -- A Case Study in Mental Health
TimeSRL uses semantic abstractions from time-series data optimized via reinforcement learning to achieve better cross-dataset generalization than standard ML or LLM baselines in mental health prediction.
-
Reinforcing Human Behavior Simulation via Verbal Feedback
DITTO uses RL with verbal feedback to train LLMs for human behavior simulation, reporting 36% average gains over base models and outperforming GPT-5.4 on 6 of 10 SOUL benchmark tasks.
-
Spectral Unforgetting: Post-Hoc Recovery of Damaged Capabilities Without Retraining
DG-Hard uses Donoho-Gavish hard thresholding on the fine-tuning weight delta to separate task-aligned signal from noise-like residual, recovering damaged capabilities while preserving target-task gains.
-
ClaimDiff-RL: Fine-Grained Caption Reinforcement Learning through Visual Claim Comparison
ClaimDiff-RL introduces reference-conditioned atomic claim differences verified by a multimodal judge as the reward signal for fine-grained RL in long-form image captioning.
-
Resolving Action Bottleneck: Agentic Reinforcement Learning Informed by Token-Level Energy
ActFocus resolves the action bottleneck in agentic RL by reweighting token gradients toward action tokens using observed reward variance and an energy-based uncertainty term, outperforming PPO and GRPO by up to 65 percentage points.
-
Early Data Exposure Improves Robustness to Subsequent Fine-Tuning
Early mixing of post-training data into pretraining improves retention of acquired capabilities after subsequent fine-tuning in language models.
-
Trust the Batch, On- or Off-Policy: Adaptive Policy Optimization for RL Post-Training
A new RL objective adapts trust-region and off-policy handling automatically via normalized effective sample size of batch policy ratios, matching tuned baselines without new hyperparameters.
-
Rotation-Preserving Supervised Fine-Tuning
RPSFT improves the in-domain versus out-of-domain performance trade-off during LLM supervised fine-tuning by penalizing rotations in pretrained singular subspaces as a proxy for loss-sensitive directions.
-
Experience Sharing in Mutual Reinforcement Learning for Heterogeneous Language Models
Mutual Reinforcement Learning allows heterogeneous LLMs to exchange experience through mechanisms like Peer Rollout Pooling, Cross-Policy GRPO Advantage Sharing, and Success-Gated Transfer, with outcome-level sharing identified as favorable on the stability-support trade-off.
-
Why Does Agentic Safety Fail to Generalize Across Tasks?
Agentic safety fails to generalize across tasks because the task-to-safe-controller mapping has a higher Lipschitz constant than the task-to-controller mapping alone, as proven in linear-quadratic control and demonstrated in quadcopter and LLM experiments.
-
Power Distribution Bridges Sampling, Self-Reward RL, and Self-Distillation
The power distribution is the target of power sampling, the closed-form solution to self-reward KL-regularized RL, and the basis for power self-distillation that matches sampling performance at lower cost.
-
A Meta Reinforcement Learning Approach to Goals-Based Wealth Management
MetaRL pre-trained on GBWM problems delivers near-optimal dynamic strategies in 0.01s achieving 97.8% of DP optimal utility and handles larger problems where DP fails.
-
Diversity in Large Language Models under Supervised Fine-Tuning
TOFU loss mitigates the narrowing of generative diversity in LLMs after supervised fine-tuning by addressing neglect of low-frequency patterns and forgetting of prior knowledge.
-
Dictionary-Aligned Concept Control for Safeguarding Multimodal LLMs
DACO curates a 15,000-concept dictionary from 400K image-caption pairs and uses it to initialize an SAE that enables granular, concept-specific steering of MLLM activations, raising safety scores on MM-SafetyBench and JailBreakV while preserving general capabilities.
-
Hindsight-Anchored Policy Optimization: Turning Failure into Feedback in Sparse Reward Settings
HAPO adds a hindsight-anchored SSI operator with Thompson gating to GRPO-style RLVR, achieving asymptotic consistency that recovers unbiased on-policy gradients as the policy improves.
-
On the optimization dynamics of RLVR: Gradient gap and step size thresholds
The paper defines a Gradient Gap for RLVR policy gradients and proves a sharp step-size threshold below which training converges and above which it collapses, with predictions for length and success-rate scaling validated in simulations and on Qwen2.5-Math-7B.
-
Improving Inverse Folding for Peptide Design with Diversity-regularized Direct Preference Optimization
Diversity-regularized DPO fine-tuning of ProteinMPNN improves structural similarity scores by at least 8% over base model and sequence diversity by up to 20% over standard DPO for peptide inverse folding on OpenFold structures.
-
Zephyr: Direct Distillation of LM Alignment
Zephyr-7B achieves state-of-the-art chat benchmark results among 7B models by distilling alignment via dDPO on AI feedback preferences, surpassing the 70B Llama-2-Chat model on MT-Bench with no human data required.
-
Jailbreaking Black Box Large Language Models in Twenty Queries
PAIR uses an attacker LLM to iteratively craft effective jailbreak prompts for black-box target LLMs in fewer than 20 queries.
-
SmoothLLM: Defending Large Language Models Against Jailbreaking Attacks
SmoothLLM mitigates jailbreaking attacks on LLMs by randomly perturbing multiple copies of a prompt at the character level and aggregating the outputs to detect adversarial inputs.
-
Training Diffusion Models with Reinforcement Learning
DDPO uses policy gradients on the denoising process to optimize diffusion models for arbitrary rewards like human feedback or compressibility.
-
Aligning Text-to-Image Models using Human Feedback
A three-stage fine-tuning process uses human ratings to train a reward model and then improves text-to-image alignment by maximizing reward-weighted likelihood.
-
Solving math word problems with process- and outcome-based feedback
On GSM8K, outcome-based supervision achieves similar final-answer error rates to process-based with less labeling, but process-based or learned reward models are needed to reach 3.4% reasoning error among correct solutions.
-
Large Language Models Are Human-Level Prompt Engineers
APE generates instruction candidates via LLM and selects the best by zero-shot performance of a second LLM, matching or beating human prompts on 19 of 24 NLP tasks.
-
Scaling Laws and Interpretability of Learning from Repeated Data
Repeating 0.1% of training data 100 times degrades an 800M parameter model's performance to that of a 400M model by damaging copying mechanisms and induction heads associated with generalization.
-
Constitutional On-Policy Safe Distillation
COPSD uses a Cross-SFT cold-start followed by constitution-conditioned distillation to achieve stronger safety-helpfulness balance and lower safety tax on reasoning than prior on-policy self-distillation methods.
-
Automating Formal Verification with Reinforcement Learning and Recursive Inference
RLVR training raises verified Dafny pass rates from 9.7% to 31.1% on a filtered benchmark while a Lean proof scaffold lifts success from 46.2% to 69.2% on a pilot set and solves 7 of 42 prior unsolved tasks.