Sliding-window transformers without positional encodings are Turing complete because the sliding window breaks permutation symmetry and suffices to simulate Post machines via a constant-size histogram state.
hub Canonical reference
Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs
Canonical reference. 93% of citing Pith papers cite this work as background.
63
Pith papers citing it
Background
93% of classified citations
abstract
The remarkable performance of models like the OpenAI o1 can be attributed to their ability to emulate human-like long-time thinking during inference. These models employ extended chain-of-thought (CoT) processes, exploring multiple strategies to enhance problem-solving capabilities. However, a critical question remains: How to intelligently and efficiently scale computational resources during testing. This paper presents the first comprehensive study on the prevalent issue of overthinking in these models, where excessive computational resources are allocated for simple problems with minimal benefit. We introduce novel efficiency metrics from both outcome and process perspectives to evaluate the rational use of computational resources by o1-like models. Using a self-training paradigm, we propose strategies to mitigate overthinking, streamlining reasoning processes without compromising accuracy. Experimental results show that our approach successfully reduces computational overhead while preserving model performance across a range of testsets with varying difficulty levels, such as GSM8K, MATH500, GPQA, and AIME.
hub tools
citation-role summary
background 15
citation-polarity summary
roles
background 15representative citing papers
DeepMath-103K is a new 103K-problem mathematical dataset with high difficulty, rigorous decontamination, and verifiable answers to support RL training of language-model reasoning.
DASH assigns segment-level credit in reasoning traces using drift toward ground-truth answers, yielding 50.8% accuracy on AIME25 versus 45.4% for GRPO while reducing overthinking behaviors.
ReSum trains LLMs via RLVR to self-summarize reasoning trajectories, yielding 4% average performance gains and 18.6% shorter rollouts through contrastive rollout branches.
SWITCH uses explicit <swi> and </swi> boundary tokens to make latent chain-of-thought compatible with on-policy RL (GRPO) and open to causal mechanistic probing, outperforming prior hidden-state recurrence methods.
KCSAT-ML benchmark supplies human error rates for math problems and DRG metric exposes that model accuracy collapses on high-human-error items while test-time scaling shows non-monotonic gains and alignment failures.
KnotBench benchmark shows state-of-the-art VLMs perform near random on diagrammatic knot reasoning tasks and lack ability to simulate structural moves.
LEAD uses online adaptive mechanisms including Potential-Scaled Instability and symmetric efficiency rewards based on correct rollouts to achieve higher accuracy-efficiency scores with substantially shorter reasoning outputs than base models on math benchmarks.
Post-Reasoning boosts LLM accuracy by reversing the usual answer-after-reasoning order, delivering mean relative gains of 17.37% across 117 model-benchmark pairs with zero extra cost.
Large Reasoning Models override their own initial safety recognition during multi-step reasoning in a failure mode called Self-Jailbreak, which Chain-of-Guardrail mitigates through targeted trajectory-level step interventions.
Evaluations of 53 LLMs on 14 basic math tasks show reasoning models use ~18x more tokens with sometimes lower accuracy, non-monotonic gains from extended budgets, and sharp performance drops under token constraints.
LCPO trains L1 reasoning models to adhere to prompt-specified CoT lengths, supporting accuracy-compute trade-offs and yielding short reasoning models that outperform larger baselines at matched lengths.
Adversaries can use crafted scene text to trigger overthinking in LVLM-based robots, producing transferable slowdowns up to 6.96x latency amplification.
SEAR trains one LLM via adversarial process rewards to explore harmful reasoning paths but flip to safe outputs, reducing over-refusal while preserving safety.
Epi2Diff extracts cognitive episode sequences from LRM reasoning traces and combines them with semantic features to predict human item difficulty, outperforming baselines on four educational datasets.
DyCon dynamically controls reasoning depth in LRMs by modeling evolving difficulty from step-level embeddings, reducing redundant steps across multiple benchmarks.
ThoughtFold applies introspective redundancy detection within correct CoT trajectories to create sub-trajectory spectra, then uses masked preference optimization to penalize redundant explorations, yielding 56% token reduction on DeepSeek-R1-Distill-Qwen-7B while preserving accuracy.
ALAR trains LLM agents to perform most reasoning in a latent space supervised by actions and escalates to explicit CoT only when needed, cutting tokens by up to 84.6% while preserving accuracy on search and tool-use benchmarks.
Post-training quantization increases overthinking errors in reasoning models; a logit penalty on curated overthinking markers reduces CoT length 12-23% without accuracy loss.
SR²AM achieves competitive Pass@1 accuracy on diverse tasks with 25.8-95.3% fewer reasoning tokens than much larger models by using self-regulated simulative planning trained via supervised learning and RL.
CES applies conditional bidirectional entropy control on top of DAPO to improve accuracy and shorten responses on mathematical benchmarks for 7B and 1.5B LLMs.
BET reduces reasoning tokens by about 55% on average while improving performance across benchmarks by learning to short-solve easy queries, fold early on unsolvable ones, and preserve budget for hard solvable queries.
SPEX delivers 1.2-3x speedup on ToT algorithms via speculative path selection, dynamic budget allocation, and adaptive early termination, reaching up to 4.1x when combined with token-level speculative decoding.
VPG-EA applies variational posterior guidance and efficiency-aware distillation to compress LLM reasoning chains while preserving performance.
citing papers explorer
-
DeepMath-103K: A Large-Scale, Challenging, Decontaminated, and Verifiable Mathematical Dataset for Advancing Reasoning
DeepMath-103K is a new 103K-problem mathematical dataset with high difficulty, rigorous decontamination, and verifiable answers to support RL training of language-model reasoning.
-
When Models Outthink Their Safety: Unveiling and Mitigating Self-Jailbreak in Large Reasoning Models
Large Reasoning Models override their own initial safety recognition during multi-step reasoning in a failure mode called Self-Jailbreak, which Chain-of-Guardrail mitigates through targeted trajectory-level step interventions.
-
Do LLMs Overthink Basic Math Reasoning? Benchmarking the Accuracy-Efficiency Tradeoff in Language Models
Evaluations of 53 LLMs on 14 basic math tasks show reasoning models use ~18x more tokens with sometimes lower accuracy, non-monotonic gains from extended budgets, and sharp performance drops under token constraints.
-
L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning
LCPO trains L1 reasoning models to adhere to prompt-specified CoT lengths, supporting accuracy-compute trade-offs and yielding short reasoning models that outperform larger baselines at matched lengths.
-
Don't Pass@k: A Bayesian Framework for Large Language Model Evaluation
A Dirichlet-prior Bayesian estimator for model success probability replaces Pass@k, delivering faster-converging and more stable rankings with credible intervals on math benchmarks.
-
Entropy After </Think> for reasoning model early exiting
Entropy After </Think> (EAT) enables early exiting in reasoning LLMs by tracking entropy stabilization after a </think> token, cutting token use 12-22% on MATH500 and AIME2025 with no accuracy loss.
-
Thinking Sparks!: Emergent Attention Heads in Reasoning Models During Post Training
Post-training on reasoning tasks sparks the emergence of specialized attention heads that enable structured computation, with SFT adding stable heads while GRPO uses dynamic activation and pruning tied to reward signals, and controllable think models relying on compensatory heads instead of specific
-
GrACE: A Generative Approach to Better Confidence Elicitation and Efficient Test-Time Scaling in Large Language Models
GrACE is a fine-tuned generative method that uses similarity to a special token embedding for real-time calibrated confidence in LLMs and enables efficient confidence-based test-time scaling.
-
The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity
LRMs exhibit complete accuracy collapse beyond certain puzzle complexities, with reasoning effort rising then declining, outperforming standard LLMs only on medium-complexity tasks.
-
RetroInfer: A Vector Storage Engine for Scalable Long-Context LLM Inference
RetroInfer introduces the wave index and wave buffer to realize sparse KV-cache attention for long-context LLM inference with up to 4.4X throughput gains while matching full-attention accuracy.
-
Muon is Scalable for LLM Training
Muon optimizer with weight decay and update scaling achieves ~2x efficiency over AdamW for large LLMs, shown via the Moonlight 3B/16B MoE model trained on 5.7T tokens.
-
Investigating Thinking Behaviours of Reasoning-Based Language Models for Social Bias Mitigation
Reasoning LLMs aggregate social biases through stereotype repetition and irrelevant information injection in their thinking processes, and a self-review prompt mitigates this on BBQ, StereoSet, and BOLD benchmarks.
-
Early Stopping Chain-of-thoughts in Large Language Models
ES-CoT shortens LLM chain-of-thought generation by tracking runs of identical step answers after linguistic markers, cutting tokens 16% on average while keeping accuracy comparable to full CoT across six datasets and three models.
-
Pruning Long Chain-of-Thought of Large Reasoning Models via Small-Scale Preference Optimization
LCPO reduces average LRM output length by over 50% across benchmarks via targeted preference optimization while preserving reasoning performance.
-
ReasonCache: Accelerating Large Reasoning Model Serving through KV Cache Sharing
ReasonCache reuses similar KV cache states across reasoning steps in LRMs via collaborative filtering to boost serving throughput by up to 89.2% while preserving accuracy.
-
Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models
A survey organizing techniques to achieve efficient reasoning in LLMs by shortening chain-of-thought outputs.
-
Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Models
The paper unifies perspectives on Long CoT in reasoning LLMs by introducing a taxonomy, detailing characteristics of deep reasoning and reflection, and discussing emergence phenomena and future directions.
-
What Factors Affect LLMs and RLLMs in Financial Question Answering?
Prompting and agent methods boost standard LLMs on financial QA by simulating long chain-of-thought reasoning, but reasoning LLMs already have this capability and show limited further gains, while multilingual alignment helps mainly by lengthening reasoning with minimal benefit for reasoning models.
-
A Survey of Reinforcement Learning for Large Reasoning Models
A survey compiling RL methods, challenges, data resources, and applications for enhancing reasoning in large language models and large reasoning models since DeepSeek-R1.
-
A Survey of Scaling in Large Language Model Reasoning
A survey categorizing scaling in LLM reasoning across input size, steps, rounds, training, and future directions, noting that scaling can negatively affect performance.
-
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.