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.
Mixed citations
Title resolution pending
Mixed citation behavior. Most common role is background (60%).
55
Pith papers citing it
Background
60% of classified citations
citation-role summary
background 4
dataset 1
citation-polarity summary
representative citing papers
The AI Scientist framework enables LLMs to independently conduct the full scientific process from idea generation to paper writing and review, demonstrated across three ML subfields with papers costing under $15 each.
UrduMMLU is a new native-source MCQ benchmark for Urdu that reveals top LLMs reach only ~90% accuracy with large gaps on region-specific humanities content.
Inducing artificial uncertainty on trivial tasks allows training probes that achieve higher calibration on hard data than standard approaches while retaining performance on easy data.
CommonWhy is a new dataset of 15,000 why-questions for evaluating LLMs on entity-based causal commonsense reasoning grounded in Wikidata.
OSCAR reduces hallucinations in diffusion language models by localizing commitment uncertainty with cross-chain entropy on parallel trajectories and applying evidence-guided remasking.
CODI compresses explicit CoT into continuous space via self-distillation and is the first implicit method to match explicit CoT performance on GSM8k at GPT-2 scale with 3.1x compression and 28.2% higher accuracy than prior implicit approaches.
KV cache compression causes task-dependent degradation in high-density reasoning due to disrupted CoT links; ShotKV mitigates this by preserving few-shot examples as indivisible semantic units through phase separation, delivering 9-18% accuracy gains and 11% latency reduction.
LLM unlearning is reframed as inadvertently installing backdoor triggers on forget-tokens; Random Noise Augmentation is introduced as a defense that improves robustness with theoretical guarantees.
DeepSeek-V2 delivers top-tier open-source LLM performance using only 21B active parameters by compressing the KV cache 93.3% and cutting training costs 42.5% via MLA and DeepSeekMoE.
GAIA benchmark shows humans at 92% accuracy on simple real-world questions far outperform current AI systems at 15%, proposing this gap as a key milestone for general AI.
ConSA learns FA/SWA allocation via L0 masks and augmented Lagrangian constraints, outperforming rule-based baselines on 0.6B and 1.7B models with consistent layer patterns.
Manifold Power Iteration aligns MoE router rows with principal singular directions of experts via a power-then-retract process, with theory showing convergence and experiments on 1B-11B models showing gains.
TN-gram replaces per-order hash tables in n-gram memory modules with a CP tensor factorization that shares token-position factors and uses order-absorption vectors, achieving comparable or better performance with fewer parameters.
Post-hoc model-based compression of reasoning traces cuts training tokens to 12-30% and speeds training 2-7.6x while retaining up to 96% of raw-trace accuracy, though raw traces remain superior at every scale.
CaliDist calibrates LLMs by scaling confidence according to how much predictions change under semantic distractors, cutting average ECE from 23% to 7% on seven NLU benchmarks across six models.
GAIS synthesizes diverse, high-fidelity agentic tasks from real-world MCP servers and adversarial planning, outperforming LLM-only baselines on BFCL, τ²-Bench, and ACEBench with greater data efficiency.
Proxy metrics from next-token distributions over expert solutions outperform loss and compute baselines for ranking LLMs, selecting pretraining data, and extrapolating performance across compute scales.
OP-Mix is an on-policy data mixing method that uses low-rank adapter interpolation to find near-optimal data mixtures throughout language model training with reduced compute.
VecCISC filters equivalent, degenerate, or hallucinated reasoning traces via semantic clustering before critic evaluation, reducing token use by 47% with no loss in accuracy versus standard CISC.
SAMoRA is a parameter-efficient fine-tuning framework that uses semantic-aware routing and task-adaptive scaling within a Mixture of LoRA Experts to improve multi-task performance and generalization over prior methods.
HEAL mitigates entropy collapse in few-shot RLVR by selectively adding general-domain data and aligning trajectory-level entropy dynamics, matching full-shot performance with 32 target samples.
SkillFactory creates silver SFT data from a model's self-generated traces rearranged into cognitive skill formats to prime models for better skill use during subsequent RL, improving post-RL generalization and out-of-domain robustness.
Math reasoning gains in LLMs rarely transfer to general domains; RL tuning generalizes while SFT causes forgetting and representation drift.
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.
-
The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery
The AI Scientist framework enables LLMs to independently conduct the full scientific process from idea generation to paper writing and review, demonstrated across three ML subfields with papers costing under $15 each.
-
UrduMMLU: A Massive Multitask Benchmark for Urdu Language Understanding
UrduMMLU is a new native-source MCQ benchmark for Urdu that reveals top LLMs reach only ~90% accuracy with large gaps on region-specific humanities content.
-
Inducing Artificial Uncertainty in Language Models
Inducing artificial uncertainty on trivial tasks allows training probes that achieve higher calibration on hard data than standard approaches while retaining performance on easy data.
-
CommonWhy: A Dataset for Evaluating Entity-Based Causal Commonsense Reasoning in Large Language Models
CommonWhy is a new dataset of 15,000 why-questions for evaluating LLMs on entity-based causal commonsense reasoning grounded in Wikidata.
-
OSCAR: Orchestrated Self-verification and Cross-path Refinement
OSCAR reduces hallucinations in diffusion language models by localizing commitment uncertainty with cross-chain entropy on parallel trajectories and applying evidence-guided remasking.
-
CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillation
CODI compresses explicit CoT into continuous space via self-distillation and is the first implicit method to match explicit CoT performance on GSM8k at GPT-2 scale with 3.1x compression and 28.2% higher accuracy than prior implicit approaches.
-
Semantic Integrity Matters: Benchmarking and Preserving High-Density Reasoning in KV Cache Compression
KV cache compression causes task-dependent degradation in high-density reasoning due to disrupted CoT links; ShotKV mitigates this by preserving few-shot examples as indivisible semantic units through phase separation, delivering 9-18% accuracy gains and 11% latency reduction.
-
Improving LLM Unlearning Robustness via Random Perturbations
LLM unlearning is reframed as inadvertently installing backdoor triggers on forget-tokens; Random Noise Augmentation is introduced as a defense that improves robustness with theoretical guarantees.
-
DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
DeepSeek-V2 delivers top-tier open-source LLM performance using only 21B active parameters by compressing the KV cache 93.3% and cutting training costs 42.5% via MLA and DeepSeekMoE.
-
GAIA: a benchmark for General AI Assistants
GAIA benchmark shows humans at 92% accuracy on simple real-world questions far outperform current AI systems at 15%, proposing this gap as a key milestone for general AI.
-
ConSA: Controllable Sparsity in Hybrid Attention via Learnable Allocation
ConSA learns FA/SWA allocation via L0 masks and augmented Lagrangian constraints, outperforming rule-based baselines on 0.6B and 1.7B models with consistent layer patterns.
-
Redesign Mixture-of-Experts Routers with Manifold Power Iteration
Manifold Power Iteration aligns MoE router rows with principal singular directions of experts via a power-then-retract process, with theory showing convergence and experiments on 1B-11B models showing gains.
-
Tensorizing Engram: Sharing Latents Across N-Gram Embeddings is Beneficial in LLMs
TN-gram replaces per-order hash tables in n-gram memory modules with a CP tensor factorization that shares token-position factors and uses order-absorption vectors, achieving comparable or better performance with fewer parameters.
-
Compress-Distill: Reasoning Trace Compression for Efficient Knowledge Distillation
Post-hoc model-based compression of reasoning traces cuts training tokens to 12-30% and speeds training 2-7.6x while retaining up to 96% of raw-trace accuracy, though raw traces remain superior at every scale.
-
CaliDist: Calibrating Large Language Models via Behavioral Robustness to Distraction
CaliDist calibrates LLMs by scaling confidence according to how much predictions change under semantic distractors, cutting average ECE from 23% to 7% on seven NLU benchmarks across six models.
-
Scaling Agentic Capabilities via Grounded Interaction Synthesis
GAIS synthesizes diverse, high-fidelity agentic tasks from real-world MCP servers and adversarial planning, outperforming LLM-only baselines on BFCL, τ²-Bench, and ACEBench with greater data efficiency.
-
Forecasting Downstream Performance of LLMs With Proxy Metrics
Proxy metrics from next-token distributions over expert solutions outperform loss and compute baselines for ranking LLMs, selecting pretraining data, and extrapolating performance across compute scales.
-
Always Learning, Always Mixing: Efficient and Simple Data Mixing All The Time
OP-Mix is an on-policy data mixing method that uses low-rank adapter interpolation to find near-optimal data mixtures throughout language model training with reduced compute.
-
VecCISC: Improving Confidence-Informed Self-Consistency with Reasoning Trace Clustering and Candidate Answer Selection
VecCISC filters equivalent, degenerate, or hallucinated reasoning traces via semantic clustering before critic evaluation, reducing token use by 47% with no loss in accuracy versus standard CISC.
-
SAMoRA: Semantic-Aware Mixture of LoRA Experts for Task-Adaptive Learning
SAMoRA is a parameter-efficient fine-tuning framework that uses semantic-aware routing and task-adaptive scaling within a Mixture of LoRA Experts to improve multi-task performance and generalization over prior methods.
-
HEALing Entropy Collapse: Enhancing Exploration in Few-Shot RLVR via Hybrid-Domain Entropy Dynamics Alignment
HEAL mitigates entropy collapse in few-shot RLVR by selectively adding general-domain data and aligning trajectory-level entropy dynamics, matching full-shot performance with 32 target samples.
-
SkillFactory: Self-Distillation For Learning Cognitive Behaviors
SkillFactory creates silver SFT data from a model's self-generated traces rearranged into cognitive skill formats to prime models for better skill use during subsequent RL, improving post-RL generalization and out-of-domain robustness.
-
Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning
Math reasoning gains in LLMs rarely transfer to general domains; RL tuning generalizes while SFT causes forgetting and representation drift.
-
The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale
FineWeb is a curated 15T-token web dataset that produces stronger LLMs than prior open collections, while its educational subset sharply improves performance on MMLU and ARC benchmarks.
-
Identifying the Achilles' Heel: An Iterative Method for Dynamically Uncovering Factual Errors in Large Language Models
HalluHunter is a knowledge-graph and rule-based NLP framework that iteratively generates single- and multi-hop questions to uncover factual errors in LLMs, triggering errors in up to 55% of cases on nine models while preserving coverage.
-
Representation Engineering: A Top-Down Approach to AI Transparency
Representation engineering uses population-level representations in deep neural networks to monitor and manipulate cognitive phenomena like honesty and harmlessness, providing simple effective baselines for LLM safety.
-
Chain-of-Verification Reduces Hallucination in Large Language Models
Chain-of-Verification reduces hallucinations in large language models by drafting responses, planning independent verification questions, answering them separately, and generating a final verified output.
-
WizardMath: Empowering Mathematical Reasoning for Large Language Models via Reinforced Evol-Instruct
WizardMath applies RLEIF to produce open-source LLMs that reach new state-of-the-art math reasoning scores on GSM8k and MATH, with the 70B variant surpassing GPT-3.5-Turbo, Claude 2, Gemini Pro, and early GPT-4.
-
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.
-
Distilling Step-by-Step! Outperforming Larger Language Models with Less Training Data and Smaller Model Sizes
Distilling step-by-step uses LLM-generated rationales as additional supervision in a multi-task framework so that 770M-parameter models outperform 540B-parameter models on NLP benchmarks with only 80% of the data.
-
BloombergGPT: A Large Language Model for Finance
BloombergGPT is a 50B parameter LLM trained on a 708B token mixed financial and general dataset that outperforms prior models on financial benchmarks while preserving general LLM performance.
-
Automatic Chain of Thought Prompting in Large Language Models
Auto-CoT automatically builds chain-of-thought demonstrations by sampling diverse questions and letting the LLM generate reasoning chains, matching manual CoT performance on ten reasoning tasks with GPT-3.
-
PaLM: Scaling Language Modeling with Pathways
PaLM 540B demonstrates continued scaling benefits by setting new few-shot SOTA results on hundreds of benchmarks and outperforming humans on BIG-bench.
-
Matching Tasks to Objectives: Fine-Tuning and Prompt-Tuning Strategies for Encoder-Decoder Pre-trained Language Models
Introduces MTO framework for matching tasks to pre-training objectives in encoder-decoder models, achieving over 120% performance gains in few-shot commonsense tasks.
-
GRASP: Geometry-aware Residual Alignment for Scalable Pretraining Data Attribution
GRASP is a scalable method for subset-level data attribution in pretraining that models interactions via a geometry-aware quadratic penalty and claims to double rank correlation while cutting costs.
-
PC Layer: Polynomial Weight Preconditioning for Improving LLM Pre-Training
A polynomial preconditioning layer controls singular value spectra of transformer weights to stabilize pre-training, shown effective on Llama-1B and supported by convergence theory for deep linear networks.
-
Dive into Ambiguity: A*-Inspired Multi-Agents Commonsense Obfuscation Attack on LLM Prompts
An A*-inspired multi-agent framework with hierarchical rewriting and a dynamic gamma parameter generates obfuscated prompts that achieve higher LLM attack success rates with fewer attempts than exhaustive search.
-
Position: Uncertainty Quantification in LLMs is Just Unsupervised Clustering
Mainstream UQ for LLMs reduces to unsupervised clustering of internal generation consistency and therefore cannot detect confident hallucinations or provide reliable safety signals.
-
Context Convergence Improves Answering Inferential Questions
Passages made from high-convergence sentences improve LLM performance on inferential questions compared to cosine similarity selection.
-
Marco-MoE: Open Multilingual Mixture-of-Expert Language Models with Efficient Upcycling
Marco-MoE delivers open multilingual MoE models with 5% activation sparsity that outperform similarly sized dense models on English and multilingual benchmarks through efficient upcycling.
-
Commonsense Knowledge with Negation: A Resource to Enhance Negation Understanding
Augmenting commonsense knowledge corpora with negation produces over 2M new triples that benefit LLM negation understanding when used for pre-training.
-
Towards Scalable Lifelong Knowledge Editing with Selective Knowledge Suppression
LightEdit enables scalable lifelong knowledge editing in LLMs via selective knowledge retrieval and probability suppression during decoding, outperforming prior methods on ZSRE, Counterfact, and RIPE while reducing training costs.
-
Understanding the Prompt Sensitivity
LLMs disperse meaning-preserving prompts internally instead of clustering them, which produces an excessively high upper bound on output log-probability differences via Taylor expansion and Cauchy-Schwarz.
-
Learning Uncertainty from Sequential Internal Dispersion in Large Language Models
SIVR detects LLM hallucinations by learning from token-wise and layer-wise variance patterns in internal hidden states, outperforming baselines with better generalization and less training data.
-
TELL-TALE: Task Efficient LLMs with Task Aware Layer Elimination
TALE selectively prunes task-detrimental layers in LLMs at inference time to match or exceed baseline performance with lower computational cost across multiple models and tasks.
-
SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model
SmolLM2 is a 1.7B-parameter language model that outperforms Qwen2.5-1.5B and Llama3.2-1B after overtraining on 11 trillion tokens using custom FineMath, Stack-Edu, and SmolTalk datasets in a multi-stage pipeline.
-
PaLM 2 Technical Report
PaLM 2 reports state-of-the-art results on language, reasoning, and multilingual tasks with improved efficiency over PaLM.
-
Hybrid Adversarial Defence for Natural Language Understanding Tasks
Hybrid entropy-uncertainty-geometric defence improves clean accuracy by up to 43% and adversarial robustness by up to 65% on NLU and security benchmarks.
-
Benchmarked Yet Not Measured -- Generative AI Should be Evaluated Against Real-World Utility
Generative AI evaluation must shift from static benchmark scores to measuring sustained improvements in human capabilities within specific deployment contexts.