TIDE enables the first cross-architecture distillation of dLLMs, improving a 0.6B student by 1.53 average points over baselines when trained from 8B dense and 16B MoE teachers.
super hub Mixed citations
write newline
Mixed citation behavior. Most common role is unclear (62%).
301
Pith papers citing it
unclear
62% of classified citations
hub tools
citation-role summary
background 8
other 4
method 1
citation-polarity summary
claims ledger
- background Table A1: Comparison of BAS for frontier models across tasks when varying the risk-prior w(t). Higher scores indicate better alignment with expressed uncertainty. The standardBAS (Uniform: w(t) = 1) serves as the baseline, while Linear and Quadratic weights simulate increasingly safety-critical environments. Identical ECE, different BAS.Consider two models evaluated on four examples with correctness labelsZ= [1, 1, 0, 0]. The models produce the following confidence values: Example 1 2 3 4 Z1 1 0
authors
co-cited works
representative citing papers
JumpLoRA uses JumpReLU gating to induce adaptive sparsity in LoRA blocks, achieving dynamic parameter isolation that prevents task interference and improves continual learning performance over IncLoRA and ELLA.
LLM judges exhibit up to 9.8 percentage point leniency bias from stakes signaling in prompts, acting implicitly without mentioning it in chain-of-thought.
InfiniteScienceGym procedurally generates unbounded scientific repositories with exact ground-truth QA pairs to benchmark LLMs on data reasoning, abstention, and tool use without static datasets.
EnsembleCert and ScaLabelCert enable tighter and exact certificates for neural network robustness against label-flipping attacks by leveraging white-box information and neural tangent kernel equivalence.
Steered LLM activations are non-surjective: under practical assumptions, they lie outside the set of states reachable from any discrete prompt.
AgentSocialBench demonstrates that privacy preservation is fundamentally harder in human-centered agentic social networks than in single-agent cases due to cross-domain coordination pressures and an abstraction paradox where privacy instructions increase discussion of sensitive information.
MiCP is the first conformal prediction method for multi-turn LLM pipelines that allocates per-turn error budgets to enable adaptive stopping with an overall coverage guarantee, shown to reduce turns and cost on RAG and ReAct benchmarks.
The paper proves W[1]-hardness parameterized by dimension d for positivity, zonotope containment, max approximation, and L_p-Lipschitz constants in 2- and 3-layer ReLU networks, showing enumeration methods are optimal under ETH.
RLCracker is a reinforcement learning attack that erases LLM watermarks at 98.5% success rate with minimal data and generalizes across ten schemes and multiple model sizes.
Establishes an unconditional robustness threshold of 1-1/q for zero-bit tamper-detection codes in watermarking, with matching constructions and experimental confirmation on image models.
ErrorRadar is a new benchmark of 2,500 multimodal K-12 math problems for MLLM error step identification and categorization, where GPT-4o trails human experts by ~10%.
BEAVER is the first text-to-SQL benchmark from private enterprise data warehouses, revealing SOTA agentic frameworks achieve only 10.8% accuracy on complex real-world queries.
Introduces an SDE-based framework for score-based generative modeling that unifies prior methods, enables predictor-corrector sampling and neural ODE likelihoods, and achieves SOTA unconditional image generation on CIFAR-10.
A noisy top-k gated mixture-of-experts layer between LSTMs scales neural networks to 137B parameters with sub-linear compute, beating SOTA on language modeling and machine translation.
A first-order stochastic optimizer that maintains bias-corrected exponential moving averages of the gradient and its square, dividing the former by the square root of the latter to set per-parameter step sizes.
AutoSP automates sequence parallelism and long-context activation checkpointing via compilation, enabling up to 2.7x longer training contexts on NVIDIA hardware with negligible throughput loss.
C2C is a new testbed where LM agents negotiate differently from humans and targeted prompting raises their win rate from 22.2% to 32.7% across 1,100+ games.
XGRAG uses graph perturbations to quantify component contributions in GraphRAG and achieves 14.81% better explanation quality than text-based baselines on QA datasets, with correlations to graph centrality.
GraphPlanner augments multi-agent LLM routing with a heterogeneous graph memory and RL-optimized MDP workflow generation, delivering up to 9.3% higher accuracy and over 99% lower GPU cost than prior routers while supporting zero-shot generalization.
MMEB-V3 benchmark shows omni-modality embedding models fail to enforce instruction-specified modality constraints and exhibit asymmetric, query-biased retrieval.
A new SFT framework for MoE models combines bias-driven sparsification with gated condenser experts to retain long-tailed expert information, outperforming DenseMixer and ESFT by over 2.5% on math reasoning and commonsense QA benchmarks.
Pliable rejection sampling learns a kernel-based proposal to enable efficient i.i.d. sampling from target distributions f with high-probability correctness and a guarantee on accepted samples.
Stimuli with low intra-modal dispersion among vision models elicit up to twice the cross-modal alignment with language models compared to high-dispersion stimuli.
citing papers explorer
-
Turning the TIDE: Cross-Architecture Distillation for Diffusion Large Language Models
TIDE enables the first cross-architecture distillation of dLLMs, improving a 0.6B student by 1.53 average points over baselines when trained from 8B dense and 16B MoE teachers.
-
JumpLoRA: Sparse Adapters for Continual Learning in Large Language Models
JumpLoRA uses JumpReLU gating to induce adaptive sparsity in LoRA blocks, achieving dynamic parameter isolation that prevents task interference and improves continual learning performance over IncLoRA and ELLA.
-
Context Over Content: Exposing Evaluation Faking in Automated Judges
LLM judges exhibit up to 9.8 percentage point leniency bias from stakes signaling in prompts, acting implicitly without mentioning it in chain-of-thought.
-
InfiniteScienceGym: An Unbounded, Procedurally-Generated Benchmark for Scientific Analysis
InfiniteScienceGym procedurally generates unbounded scientific repositories with exact ground-truth QA pairs to benchmark LLMs on data reasoning, abstention, and tool use without static datasets.
-
Exact Certification of Neural Networks and Partition Aggregation Ensembles against Label Poisoning
EnsembleCert and ScaLabelCert enable tighter and exact certificates for neural network robustness against label-flipping attacks by leveraging white-box information and neural tangent kernel equivalence.
-
Steered LLM Activations are Non-Surjective
Steered LLM activations are non-surjective: under practical assumptions, they lie outside the set of states reachable from any discrete prompt.
-
AgentSocialBench: Evaluating Privacy Risks in Human-Centered Agentic Social Networks
AgentSocialBench demonstrates that privacy preservation is fundamentally harder in human-centered agentic social networks than in single-agent cases due to cross-domain coordination pressures and an abstraction paradox where privacy instructions increase discussion of sensitive information.
-
Adaptive Stopping for Multi-Turn LLM Reasoning
MiCP is the first conformal prediction method for multi-turn LLM pipelines that allocates per-turn error budgets to enable adaptive stopping with an overall coverage guarantee, shown to reduce turns and cost on RAG and ReAct benchmarks.
-
Parameterized Hardness of Zonotope Containment and Neural Network Verification
The paper proves W[1]-hardness parameterized by dimension d for positivity, zonotope containment, max approximation, and L_p-Lipschitz constants in 2- and 3-layer ReLU networks, showing enumeration methods are optimal under ETH.
-
RLCracker: Evaluating the Worst-Case Vulnerability of LLM Watermarks with Adaptive RL Attacks
RLCracker is a reinforcement learning attack that erases LLM watermarks at 98.5% success rate with minimal data and generalizes across ten schemes and multiple model sizes.
-
The Coding Limits of Robust Watermarking for Generative Models
Establishes an unconditional robustness threshold of 1-1/q for zero-bit tamper-detection codes in watermarking, with matching constructions and experimental confirmation on image models.
-
ErrorRadar: Benchmarking Complex Mathematical Reasoning of Multimodal Large Language Models Via Error Detection
ErrorRadar is a new benchmark of 2,500 multimodal K-12 math problems for MLLM error step identification and categorization, where GPT-4o trails human experts by ~10%.
-
BEAVER: An Enterprise Benchmark for Text-to-SQL
BEAVER is the first text-to-SQL benchmark from private enterprise data warehouses, revealing SOTA agentic frameworks achieve only 10.8% accuracy on complex real-world queries.
-
Score-Based Generative Modeling through Stochastic Differential Equations
Introduces an SDE-based framework for score-based generative modeling that unifies prior methods, enables predictor-corrector sampling and neural ODE likelihoods, and achieves SOTA unconditional image generation on CIFAR-10.
-
Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer
A noisy top-k gated mixture-of-experts layer between LSTMs scales neural networks to 137B parameters with sub-linear compute, beating SOTA on language modeling and machine translation.
-
Adam: A Method for Stochastic Optimization
A first-order stochastic optimizer that maintains bias-corrected exponential moving averages of the gradient and its square, dividing the former by the square root of the latter to set per-parameter step sizes.
-
AutoSP: Unlocking Long-Context LLM Training Via Compiler-Based Sequence Parallelism
AutoSP automates sequence parallelism and long-context activation checkpointing via compilation, enabling up to 2.7x longer training contexts on NVIDIA hardware with negligible throughput loss.
-
Cooperate to Compete: Strategic Coordination in Multi-Agent Conquest
C2C is a new testbed where LM agents negotiate differently from humans and targeted prompting raises their win rate from 22.2% to 32.7% across 1,100+ games.
-
XGRAG: A Graph-Native Framework for Explaining KG-based Retrieval-Augmented Generation
XGRAG uses graph perturbations to quantify component contributions in GraphRAG and achieves 14.81% better explanation quality than text-based baselines on QA datasets, with correlations to graph centrality.
-
GraphPlanner: Graph Memory-Augmented Agentic Routing for Multi-Agent LLMs
GraphPlanner augments multi-agent LLM routing with a heterogeneous graph memory and RL-optimized MDP workflow generation, delivering up to 9.3% higher accuracy and over 99% lower GPU cost than prior routers while supporting zero-shot generalization.
-
MMEB-V3: Measuring the Performance Gaps of Omni-Modality Embedding Models
MMEB-V3 benchmark shows omni-modality embedding models fail to enforce instruction-specified modality constraints and exhibit asymmetric, query-biased retrieval.
-
Preserving Long-Tailed Expert Information in Mixture-of-Experts Tuning
A new SFT framework for MoE models combines bias-driven sparsification with gated condenser experts to retain long-tailed expert information, outperforming DenseMixer and ESFT by over 2.5% on math reasoning and commonsense QA benchmarks.
-
Pliable rejection sampling
Pliable rejection sampling learns a kernel-based proposal to enable efficient i.i.d. sampling from target distributions f with high-probability correctness and a guarantee on accepted samples.
-
Modulating Cross-Modal Convergence with Single-Stimulus, Intra-Modal Dispersion
Stimuli with low intra-modal dispersion among vision models elicit up to twice the cross-modal alignment with language models compared to high-dispersion stimuli.
-
The Consensus Trap: Rescuing Multi-Agent LLMs from Adversarial Majorities via Token-Level Collaboration
Token-level interleaving in multi-agent LLMs allows honest agents to overpower adversarial majorities through dynamic logic chaining, unlike brittle response-level majority voting.
-
Randomized Antipodal Search Done Right for Data Pareto Improvement of LLM Unlearning
RASLIK uses randomized antipodal search on linearized influence kernels to achieve data Pareto improvement in LLM unlearning, outperforming baselines with sublinear complexity and double gains in quality and efficiency.
-
Diagnosing LLM Judge Reliability: Conformal Prediction Sets and Transitivity Violations
LLM judges display per-document transitivity violations in 33-67% of cases despite low aggregate rates, while conformal prediction set widths serve as reliable indicators of document-level difficulty with cross-judge agreement.
-
Sublinear Spectral Clustering Oracle with Little Memory
Sublinear spectral clustering oracles achieve O(n^0.01) memory construction and sublinear query time for well-clusterable graphs, with a near-optimal S*T = O~(n) memory-time trade-off.
-
A Unified Model and Document Representation for On-Device Retrieval-Augmented Generation
A single model unifies retrieval and context compression for on-device RAG via shared representations, matching traditional RAG performance at 1/10 context size with no extra storage.
-
Mamba-SSM with LLM Reasoning for Feature Selection: Faithfulness-Aware Biomarker Discovery
LLM chain-of-thought filtering of Mamba saliency features on TCGA-BRCA data produces a 17-gene set with AUC 0.927 that beats both the raw 50-gene saliency list and a 5000-gene baseline while using far fewer features, though it misses many known BRCA genes.
-
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.
-
CodeSpecBench: Benchmarking LLMs for Executable Behavioral Specification Generation
CodeSpecBench shows LLMs achieve at most 20.2% pass rate on repository-level executable behavioral specification generation, revealing that strong code generation does not imply deep semantic understanding.
-
CocoaBench: Evaluating Unified Digital Agents in the Wild
CocoaBench shows the best tested unified digital agents succeed on only 45.1% of human-designed tasks that demand integrated vision, search, and coding.
-
Shuffling the Data, Stretching the Step-size: Sharper Bias in constant step-size SGD
Combining random reshuffling and Richardson-Romberg extrapolation yields cubic bias refinement and better MSE for constant-step SGD on structured non-monotone variational inequalities.
-
Jailbreaking the Matrix: Nullspace Steering for Controlled Model Subversion
HMNS is a new jailbreak method that uses causal head identification and nullspace-constrained injection to achieve higher attack success rates than prior techniques on aligned language models.
-
Many-Tier Instruction Hierarchy in LLM Agents
ManyIH and ManyIH-Bench address instruction conflicts in LLM agents with up to 12 privilege levels across 853 tasks, revealing frontier models achieve only ~40% accuracy.
-
HiL-Bench (Human-in-Loop Benchmark): Do Agents Know When to Ask for Help?
HiL-Bench shows frontier AI agents fail to ask for help on incomplete tasks, recovering only a fraction of full-information performance, but RL training on Ask-F1 reward improves judgment and transfers across domains.
-
Wiring the 'Why': A Unified Taxonomy and Survey of Abductive Reasoning in LLMs
The paper delivers the first survey of abductive reasoning in LLMs, a unified two-stage taxonomy, a compact benchmark, and an analysis of gaps relative to deductive and inductive reasoning.
-
Multilingual Embedding Probes Fail to Generalize Across Learner Corpora
Multilingual embedding probes achieve strong in-distribution CEFR prediction (QWK ≈ 0.7) but fail to generalize across corpora, converging to uniform predictions and capturing corpus-specific features instead of language-general proficiency.
-
Learning to Interrupt in Language-based Multi-agent Communication
HANDRAISER learns optimal interruption points in multi-agent LLM communication using estimated future reward and cost, achieving 32.2% lower communication cost with comparable or better task results across games, scheduling, and debate.
-
Attention Flows: Tracing LLM Conceptual Engagement via Story Summaries
LLM novel summaries emphasize endings more than human ones, measured by aligning summary sentences to referenced chapters.
-
FrontierFinance: A Long-Horizon Computer-Use Benchmark of Real-World Financial Tasks
FrontierFinance benchmark shows human financial experts outperform state-of-the-art LLMs by achieving higher scores and more client-ready outputs on realistic long-horizon tasks.
-
$S^3$: Stratified Scaling Search for Test-Time in Diffusion Language Models
S³ is a verifier-guided stratified search over denoising trajectories that reallocates inference compute to improve output quality from fixed diffusion language models on reasoning benchmarks.
-
EffiPair: Improving the Efficiency of LLM-generated Code with Relative Contrastive Feedback
EffiPair uses relative contrastive feedback from program pairs to iteratively improve the efficiency of LLM-generated code at test time, achieving speedups with reduced overhead.
-
MMORF: A Multi-agent Framework for Designing Multi-objective Retrosynthesis Planning Systems
MMORF provides a modular multi-agent framework for multi-objective retrosynthesis planning, with MASIL and RFAS systems showing strong safety, cost, and success metrics on a new 218-task benchmark.
-
From Curiosity to Caution: Mitigating Reward Hacking for Best-of-N with Pessimism
Caution mitigates reward hacking in Best-of-N sampling by penalizing prediction errors from an error model as signals of uncertainty, with empirical improvements and provable gains over standard BoN in a linear setting.
-
Retrieval Augmented Conversational Recommendation with Reinforcement Learning
RAR retrieves candidate items from a 300k-movie corpus then uses LLM generation with RL feedback to produce context-aware recommendations that outperform baselines on benchmarks.
-
MetaSAEs: Joint Training with a Decomposability Penalty Produces More Atomic Sparse Autoencoder Latents
Joint training of a primary SAE with a meta SAE that applies a decomposability penalty on decoder directions produces more atomic latents, shown by 7.5% lower mean absolute phi and 7.6% higher fuzzing scores on GPT-2.
-
CresOWLve: Benchmarking Creative Problem-Solving Over Real-World Knowledge
CresOWLve benchmark shows frontier LLMs retrieve relevant real-world facts but struggle to form creative connections, with up to 17% lower performance on creative questions than factual ones.
-
BAS: A Decision-Theoretic Approach to Evaluating Large Language Model Confidence
BAS aggregates utility from an answer-or-abstain model across risk thresholds and is uniquely maximized by truthful confidence estimates.