Active lifting of data distributions via latent-sequence sampling and max-rate uncertainty reduction formally derives slow-thinking LLMs and places them on representation and sampler hierarchies that can be climbed.
hub Canonical reference
Quiet-STaR: Language Models Can Teach Themselves to Think Before Speaking
Canonical reference. 78% of citing Pith papers cite this work as background.
49
Pith papers citing it
13
external citations · Pith
Background
78% of classified citations
abstract
When writing and talking, people sometimes pause to think. Although reasoning-focused works have often framed reasoning as a method of answering questions or completing agentic tasks, reasoning is implicit in almost all written text. For example, this applies to the steps not stated between the lines of a proof or to the theory of mind underlying a conversation. In the Self-Taught Reasoner (STaR, Zelikman et al. 2022), useful thinking is learned by inferring rationales from few-shot examples in question-answering and learning from those that lead to a correct answer. This is a highly constrained setting -- ideally, a language model could instead learn to infer unstated rationales in arbitrary text. We present Quiet-STaR, a generalization of STaR in which LMs learn to generate rationales at each token to explain future text, improving their predictions. We address key challenges, including 1) the computational cost of generating continuations, 2) the fact that the LM does not initially know how to generate or use internal thoughts, and 3) the need to predict beyond individual next tokens. To resolve these, we propose a tokenwise parallel sampling algorithm, using learnable tokens indicating a thought's start and end, and an extended teacher-forcing technique. Encouragingly, generated rationales disproportionately help model difficult-to-predict tokens and improve the LM's ability to directly answer difficult questions. In particular, after continued pretraining of an LM on a corpus of internet text with Quiet-STaR, we find zero-shot improvements on GSM8K (5.9%$\rightarrow$10.9%) and CommonsenseQA (36.3%$\rightarrow$47.2%) and observe a perplexity improvement of difficult tokens in natural text. Crucially, these improvements require no fine-tuning on these tasks. Quiet-STaR marks a step towards LMs that can learn to reason in a more general and scalable way.
hub tools
citation-role summary
background 7
method 2
citation-polarity summary
representative citing papers
DecompRL is an RL method that learns modular code decomposition for LLMs, enabling exponential candidate generation via recombination to solve harder coding problems with lower GPU cost.
CRAFT is a three-pillar credit assignment scheme that uses counterfactual token importance from GRPO sibling rollouts to provide signed per-token distillation signals in self-distilled agentic RL.
PearlVLA achieves SOTA on LIBERO by separating VLM representations into visual grounding and an iterative latent plan branch refined via world model queries and RefineNet with process-reward RL.
CaLIR learns continuous latent intent states guided by product category hierarchies for generative retrieval, combining hierarchical reasoning and dynamic prefix tries to balance effectiveness and low-latency inference on multilingual e-commerce data.
LatentOmni proposes a latent-space cross-modal reasoning framework that uses feature-level supervision and Omni-Sync Position Embedding to align and synchronize audio-visual latents, supported by a new 35K interleaved reasoning dataset and showing gains over text CoT baselines.
PERL augments frozen CLIP with a shared recurrent reasoning module of roughly 6K parameters that iteratively refines representations via latent token injection, delivering strong base-to-novel and transfer performance across 15 benchmarks.
ATLAS uses a single functional token to unify agentic and latent visual reasoning without image generation or external execution.
Presents a likelihood-based benchmark for equation-suffix prediction in technical papers with controls to detect shortcut vulnerabilities in model forecasts.
Language models achieve a perfect LSAT score, with experiments showing that internal thinking phases and a fine-tuned process reward model are key to high performance on logical reasoning questions.
PLUME uses latent-state autoregressive rollouts and a progressive training curriculum to deliver efficient reasoning for universal multimodal embeddings without generating explicit rationales.
Laser reformulates visual reasoning via Dynamic Windowed Alignment Learning to maintain latent superposition of global features, delivering 5.03% average gains over Monet and over 97% fewer inference tokens on six benchmarks.
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.
Coconut lets LLMs perform reasoning directly in continuous latent space by recycling hidden states as inputs, outperforming standard chain-of-thought on search-intensive logical tasks with better accuracy-efficiency trade-offs.
SFT supplies entangled compositional traces of atomic skills and routing modules; RL identifies those modules and enables recombination on novel compositions outside the SFT support.
Synthetic page-ranked reasoning traces plus low-strength model merging give a 32B VLM 58.3 on MMLongBenchDoc, beating a 235B teacher while cutting output tokens ~12× versus explicit reasoning.
CoAT adds a continuous latent thinking space to LALMs via expert distillation to retain acoustic information, yielding gains on audio reasoning, understanding, music, emotion, and transcription benchmarks across three models.
LARM enables test-time compute scaling in non-autoregressive ASR via depth-conditioned looping with CTC checkpoints, supervision embeddings, FiLM conditioning, and posterior feedback, yielding lower WER on LibriSpeech with more loops.
Contribution Weights combine attention, value magnitude, and directional alignment to measure token influence more faithfully than attention alone, and show attention sinks actively suppress information via a convex sink-rate to output-norm relationship.
STRIDE co-trains generator and verifier on outcome rewards alone to deliver learnable stepwise language feedback that redirects LLM reasoning trajectories and outperforms scalar-reward baselines.
Distills 3D spatial reasoning from a 7B teacher VLM to a 2.29B student using VGGT encoder, multi-task losses, and Hidden CoT latent tokens, yielding 8.7x lower latency with 54-72% performance retention on ScanNet and 3D-FRONT.
SimpleTES scales test-time evaluation in LLMs to discover state-of-the-art solutions on 21 scientific problems across six domains, outperforming frontier models and optimization pipelines with examples like 2x faster LASSO and new Erdos constructions.
LLM agent committees exhibit representational collapse with mean cosine similarity of 0.888, and diversity-aware consensus reaches 87% accuracy on GSM8K versus 84% for self-consistency at lower cost.
CAST improves LLM stability for text summarization and tagging on tabular data by up to 16.2% via algorithmic prompting and thinking-before-speaking while preserving output quality.
citing papers explorer
-
A First-Principles Theory of Slow Thinking and Active Perception
Active lifting of data distributions via latent-sequence sampling and max-rate uncertainty reduction formally derives slow-thinking LLMs and places them on representation and sampler hierarchies that can be climbed.
-
DecompRL: Solving Harder Problems by Learning Modular Code Generation
DecompRL is an RL method that learns modular code decomposition for LLMs, enabling exponential candidate generation via recombination to solve harder coding problems with lower GPU cost.
-
CRAFT: Counterfactual Credit Assignment from Free Sibling Rollouts for Self-Distilled Agentic Reinforcement Learning
CRAFT is a three-pillar credit assignment scheme that uses counterfactual token importance from GRPO sibling rollouts to provide signed per-token distillation signals in self-distilled agentic RL.
-
PearlVLA: Progressive Embodied Action-Plan Refinement in Latent Space
PearlVLA achieves SOTA on LIBERO by separating VLM representations into visual grounding and an iterative latent plan branch refined via world model queries and RefineNet with process-reward RL.
-
Beyond Matching: Category-Guided Latent Intent Reasoning for Generative Retrieval in E-Commerce
CaLIR learns continuous latent intent states guided by product category hierarchies for generative retrieval, combining hierarchical reasoning and dynamic prefix tries to balance effectiveness and low-latency inference on multilingual e-commerce data.
-
LatentOmni: Rethinking Omni-Modal Understanding via Unified Audio-Visual Latent Reasoning
LatentOmni proposes a latent-space cross-modal reasoning framework that uses feature-level supervision and Omni-Sync Position Embedding to align and synchronize audio-visual latents, supported by a new 35K interleaved reasoning dataset and showing gains over text CoT baselines.
-
PERL: Parameter Efficient Reasoning in CLIP Latent Space
PERL augments frozen CLIP with a shared recurrent reasoning module of roughly 6K parameters that iteratively refines representations via latent token injection, delivering strong base-to-novel and transfer performance across 15 benchmarks.
-
ATLAS: Agentic or Latent Visual Reasoning? One Word is Enough for Both
ATLAS uses a single functional token to unify agentic and latent visual reasoning without image generation or external execution.
-
Likelihood scoring for continuations of mathematical text: a self-supervised benchmark with tests for shortcut vulnerabilities
Presents a likelihood-based benchmark for equation-suffix prediction in technical papers with controls to detect shortcut vulnerabilities in model forecasts.
-
AI Achieves a Perfect LSAT Score
Language models achieve a perfect LSAT score, with experiments showing that internal thinking phases and a fine-tuned process reward model are key to high performance on logical reasoning questions.
-
PLUME: Latent Reasoning Based Universal Multimodal Embedding
PLUME uses latent-state autoregressive rollouts and a progressive training curriculum to deliver efficient reasoning for universal multimodal embeddings without generating explicit rationales.
-
Forest Before Trees: Latent Superposition for Efficient Visual Reasoning
Laser reformulates visual reasoning via Dynamic Windowed Alignment Learning to maintain latent superposition of global features, delivering 5.03% average gains over Monet and over 97% fewer inference tokens on six benchmarks.
-
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.
-
Training Large Language Models to Reason in a Continuous Latent Space
Coconut lets LLMs perform reasoning directly in continuous latent space by recycling hidden states as inputs, outperforming standard chain-of-thought on search-intensive logical tasks with better accuracy-efficiency trade-offs.
-
From Reasoning Traces to Reusable Modules: Understanding Compositional Generalization in Language Model Reasoning
SFT supplies entangled compositional traces of atomic skills and routing modules; RL identifies those modules and enables recombination on novel compositions outside the SFT support.
-
Internalized Reasoning for Long-Context Visual Document Understanding
Synthetic page-ranked reasoning traces plus low-strength model merging give a 32B VLM 58.3 on MMLongBenchDoc, beating a 235B teacher while cutting output tokens ~12× versus explicit reasoning.
-
Continuous Audio Thinking for Large Audio Language Models
CoAT adds a continuous latent thinking space to LALMs via expert distillation to retain acoustic information, yielding gains on audio reasoning, understanding, music, emotion, and transcription benchmarks across three models.
-
Test-Time Compute Scaling for ASR with Depth-Conditioned Looped Transformers
LARM enables test-time compute scaling in non-autoregressive ASR via depth-conditioned looping with CTC checkpoints, supervision embeddings, FiLM conditioning, and posterior feedback, yielding lower WER on LibriSpeech with more loops.
-
Contribution Weights: A Geometrical Analysis of Self-Attention Transformers
Contribution Weights combine attention, value magnitude, and directional alignment to measure token influence more faithfully than attention alone, and show attention sinks actively suppress information via a convex sink-rate to output-norm relationship.
-
STRIDE: Learnable Stepwise Language Feedback for LLM Reasoning
STRIDE co-trains generator and verifier on outcome rewards alone to deliver learnable stepwise language feedback that redirects LLM reasoning trajectories and outperforms scalar-reward baselines.
-
Distilling 3D Spatial Reasoning into a Lightweight Vision-Language Model with CoT
Distills 3D spatial reasoning from a 7B teacher VLM to a 2.29B student using VGGT encoder, multi-task losses, and Hidden CoT latent tokens, yielding 8.7x lower latency with 54-72% performance retention on ScanNet and 3D-FRONT.
-
Evaluation-driven Scaling for Scientific Discovery
SimpleTES scales test-time evaluation in LLMs to discover state-of-the-art solutions on 21 scientific problems across six domains, outperforming frontier models and optimization pipelines with examples like 2x faster LASSO and new Erdos constructions.
-
Representational Collapse in Multi-Agent LLM Committees: Measurement and Diversity-Aware Consensus
LLM agent committees exhibit representational collapse with mean cosine similarity of 0.888, and diversity-aware consensus reaches 87% accuracy on GSM8K versus 84% for self-consistency at lower cost.
-
CAST: Achieving Stable LLM-based Text Analysis for Data Analytics
CAST improves LLM stability for text summarization and tagging on tabular data by up to 16.2% via algorithmic prompting and thinking-before-speaking while preserving output quality.
-
Compute as Teacher: Turning Inference Compute Into Reference-Free Supervision
Parallel inference rollouts aggregated into pseudo-references enable reference-free RL supervision that matches expert-annotated performance on health tasks while using 9x less test-time compute.
-
ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models
Prolonged RL training with KL control and reference policy resetting enables LLMs to develop novel reasoning strategies inaccessible to base models even under extensive sampling.
-
LIMO: Less is More for Reasoning
LIMO achieves 63.3% on AIME24 and 95.6% on MATH500 via supervised fine-tuning on roughly 1% of the data used by prior models, supporting the claim that minimal strategic examples suffice when pre-training has already encoded domain knowledge.
-
Search-o1: Agentic Search-Enhanced Large Reasoning Models
Search-o1 integrates agentic retrieval-augmented generation and a Reason-in-Documents module into large reasoning models to dynamically supply missing knowledge and improve performance on complex science, math, coding, and QA tasks.
-
MeTHanol: Modularized Thinking Language Models with Intermediate Layer Thinking, Decoding and Bootstrapping Reasoning
MeTHanol fine-tunes an intermediate LLM layer to generate thoughts in a first pass, then uses those thoughts for a second-pass answer, showing gains on Theory of Mind and vignette tasks plus adaptation to character prompts.
-
Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters
An adaptive compute-optimal strategy for scaling LLM test-time compute achieves over 4x efficiency gains versus best-of-N and lets smaller models outperform 14x larger ones on some problems.
-
Large Language Monkeys: Scaling Inference Compute with Repeated Sampling
Repeated sampling scales problem coverage log-linearly with sample count, improving SWE-bench Lite performance from 15.9% to 56% using 250 samples.
-
Artificial Intelligence for Mathematical Reasoning: An Integrated Survey of Language Models, Neuro-symbolic Systems, and Verified Discovery
AI for mathematics is best described as a supervision ladder — final answers, programs, process rewards, proof-assistant kernels — culminating in verified-discovery workflows.
-
TARPO: Token-Wise Latent-Explicit Reasoning via Action-Routing Policy Optimization
TARPO is a pure RL framework using a token-wise action router to switch between discrete token generation and latent reasoning in LLMs, with joint optimization showing outperformance on benchmarks.
-
MIRAGE: Mobile Agents with Implicit Reasoning and Generative World Models
MIRAGE compresses explicit chain-of-thought into latent vectors and adds a generative world model to predict future interface states, matching explicit reasoning performance with 3-5x fewer tokens on Android benchmarks.
-
Latent Recurrent Transformer: Architecture Exploration, Training Strategies, and Scaling Behavior
Latent Recurrent Transformer augments autoregressive transformers with a cross-layer recurrent latent pathway from prior hidden states and uses interleaved parallel training to improve loss and in-context learning at ~0.3% extra parameters.
-
Stabilizing Recurrent Dynamics for Test-Time Scalable Latent Reasoning in Looped Language Models
STARS trains looped language models with Jacobian spectral radius regularization and random loop sampling to drive latent states toward asymptotically stable fixed points, yielding reliable test-time scaling on arithmetic and mathematical reasoning tasks.
-
HypEHR: Hyperbolic Modeling of Electronic Health Records for Efficient Question Answering
A 22M-parameter hyperbolic model answers structured EHR questions with accuracy close to LLM-based systems (EHRXQA 89.5%, MIMIC-Instr 76.0%).
-
Learning to Trade Like an Expert: Cognitive Fine-Tuning for Stable Financial Reasoning in Language Models
A new fine-tuning framework with textbook-derived MCQs and simulation-based testing enables smaller open LLMs to show competitive, risk-aware financial trading behavior that outperforms baselines.
-
DRAFT: Task Decoupled Latent Reasoning for Agent Safety
DRAFT decouples agent safety judgment into latent extraction and reasoning stages, raising average benchmark accuracy from 63.27% to 91.18%.
-
Webscale-RL: Automated Data Pipeline for Scaling RL Data to Pretraining Levels
Webscale-RL generates 1.2M verifiable QA pairs from pretraining corpora, enabling RL training that matches continual pretraining performance with up to 100x fewer tokens.
-
InternVideo3: Agentify Foundation Models with Multimodal Contextual Reasoning
InternVideo3 introduces Multimodal Contextual Reasoning and M^2LA attention to enable closed-loop evidence accumulation in long-video understanding and agentic tool use, reporting strong benchmark results.
-
Shaping Schema via Language Representation as the Next Frontier for LLM Intelligence Expanding
Advanced language representations shape LLMs' schemas to improve knowledge activation and problem-solving.
-
Pangu-ACE: Adaptive Cascaded Experts for Educational Response Generation on EduBench
Pangu-ACE improves educational response quality on EduBench from 0.457 to 0.538 and format validity from 0.707 to 0.866 by routing 19.7% of samples to a 1B model while escalating the rest to 7B.
-
From System 1 to System 2: A Survey of Reasoning Large Language Models
The survey organizes the shift of LLMs toward deliberate System 2 reasoning, covering model construction techniques, performance on math and coding benchmarks, and future research directions.
-
Imitate, Explore, and Self-Improve: A Reproduction Report on Slow-thinking Reasoning Systems
STILL-2 uses imitation of distilled long-form thoughts, multi-rollout exploration on difficult problems, and iterative self-improvement of the dataset to train reasoning models that reach competitive performance on three challenging benchmarks.
-
Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems
This survey frames foundation agents using brain-inspired modular architectures and reviews challenges in evolution, collaboration, and safety.
-
The Hitchhiker's Guide to Agentic AI: From Foundations to Systems
A survey-style reference book mapping the full agentic-AI stack from transformer internals to production deployment, with no new research result.
-
Reinforcement Learning from Human Feedback
An expository book that systematically presents RLHF methods, from reward modeling to direct alignment algorithms, aimed at readers with quantitative backgrounds.
- Personalization as Inverse Planning: Learning Latent Design Intents for Agentic Slide Generation via Structural Denoising