Introduces unified span-level hallucination detection benchmark over code, tool output, and documents; fine-tuned Qwen3.5-2B reaches 0.689 span-F1 and outperforms baselines including on code-agent data.
Title resolution pending
34 Pith papers cite this work. Polarity classification is still indexing.
34
Pith papers citing it
representative citing papers
Fork-think with confidence identifies forking points via model confidence in a single path before sampling continuations, cutting tokens up to 30% and runtime up to 57% on reasoning benchmarks while matching or exceeding parallel thinking performance.
ThinkProbe builds non-generative Thought Graphs from 4200 LLM traces across 7 models and 200 questions to extract 5D cognitive profiles, finding model-level stability in reasoning structure that exceeds domain effects in four dimensions.
Agentic CLEAR automates multi-level evaluation of LLM agents, generating textual insights at system, trace, and node granularity that align with human annotations and predict task success.
A new 263k TTCW-annotated story dataset shows non-reasoning fine-tuning of Qwen3 models outperforms reasoning-supervised fine-tuning for fixed-format long-form literary review generation.
Rule2DRC is a benchmark for LLM agents synthesizing DRC scripts from natural language rules, paired with SplitTester that improves Best-of-N selection via execution-guided discriminative test generation.
Proves that RoPE attention loses locality bias and token distinction in long contexts, approaching random behavior independent of content.
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.
CircuitFormer is a 511M-parameter encoder-decoder model that generates analog circuit topologies from text prompts at 100% syntactic correctness and 83% functional success using a new subcircuit-mining tokenizer that keeps vocabulary size fixed at 512.
Retrieving structured thinking traces as a corpus improves reasoning performance on AIME, LiveCodeBench, and GPQA over standard RAG or no retrieval.
Pruning MoE LLMs according to the stationary distribution of a Markov chain over (layer, expert) routing transitions retains more task performance than local importance heuristics, with up to ~3.5% relative gains over the strongest baseline at 50% compression.
SDM sparsifies the Gated DeltaNet update rule to enable 1000x larger recurrent memory states at iso-FLOP, improving long-context recall and short-context reasoning over GDN and matching full attention at 8B scale.
Global calibration metrics like ECE are confounded by accuracy; the proposed ACE framework with three accuracy-controlled views shows many prior calibration advantages weaken or reverse.
SVR-MAD treats pre-debate signals as priors and debate results as evidence to build a sparser communication graph, cutting token use by up to 61% while preserving or raising accuracy over prior MAD methods.
SWE-Mutation benchmark shows current LLMs achieve low verification (10.20%) and detection (36.15%) rates on 2,636 mutated variants, exposing weaknesses in generating reliable test suites.
An empirical red-teaming study measures political Overton Windows across more than 30 open-source LLMs from 10 families and finds left-leaning bias, inverse size correlation, regional variation, and variable jailbreak effectiveness.
Interventions in LLM-simulated user experiments induce distribution shifts in latent attributes that create confounding bias, diagnosable with negative control outcomes and partially mitigated by adding setting-relevant persona details.
AMARIS augments rubric updates in RL for LLMs with a persistent memory of rollout analyses and prior edits, yielding gains such as +2.8 points on GPQA-Diamond over local-adaptive baselines.
Introduces predictive prefetching for RAG that anticipates retrieval needs several tokens ahead via three components, reporting up to 43.5% latency reduction and 62.4% TTFT improvement while preserving answer quality.
SP-KV trains a utility predictor jointly with the LLM to dynamically prune low-utility KV cache entries, achieving 3-10x memory reduction during generation with negligible performance loss.
BioTool dataset enables fine-tuning a 4B-parameter LLM to outperform GPT-5.1 in biomedical tool calling while improving downstream answer quality per human experts.
MultiBreak is a large diverse multi-turn jailbreak benchmark that achieves substantially higher attack success rates on LLMs than prior datasets and reveals topic-specific vulnerabilities in multi-turn settings.
Verbal-R3 uses a verbal reranker to generate analytic narratives that guide retrieval and reasoning in LLMs, achieving SOTA results on complex QA benchmarks.
Fine-tuning an LLM with DPO against a detector score plus a task metric (PUPPET) improves both the detector's AUROC and downstream task scores, beating watermarked baselines on task quality in the reported benchmarks.
citing papers explorer
-
Beyond Document Grounding: Span-Level Hallucination Detection over Code, Tool Output, and Documents
Introduces unified span-level hallucination detection benchmark over code, tool output, and documents; fine-tuned Qwen3.5-2B reaches 0.689 span-F1 and outperforms baselines including on code-agent data.
-
Fork-Think with Confidence
Fork-think with confidence identifies forking points via model confidence in a single path before sampling continuations, cutting tokens up to 30% and runtime up to 57% on reasoning benchmarks while matching or exceeding parallel thinking performance.
-
ThinkProbe: Beyond Accuracy -- Structural Profiling of Open-Ended LLM Reasoning Traces via Non-Generative Thought Graphs
ThinkProbe builds non-generative Thought Graphs from 4200 LLM traces across 7 models and 200 questions to extract 5D cognitive profiles, finding model-level stability in reasoning structure that exceeds domain effects in four dimensions.
-
Agentic CLEAR: Automating Multi-Level Evaluation of LLM Agents
Agentic CLEAR automates multi-level evaluation of LLM agents, generating textual insights at system, trace, and node granularity that align with human annotations and predict task success.
-
When Reasoning Supervision Hurts: TTCW-Based Long-Form Literary Review Generation
A new 263k TTCW-annotated story dataset shows non-reasoning fine-tuning of Qwen3 models outperforms reasoning-supervised fine-tuning for fixed-format long-form literary review generation.
-
Rule2DRC: Benchmarking LLM Agents for DRC Script Synthesis with Execution-Guided Test Generation
Rule2DRC is a benchmark for LLM agents synthesizing DRC scripts from natural language rules, paired with SplitTester that improves Best-of-N selection via execution-guided discriminative test generation.
-
RoPE Distinguishes Neither Positions Nor Tokens in Long Contexts, Provably
Proves that RoPE attention loses locality bias and token distinction in long contexts, approaching random behavior independent of content.
-
Post Reasoning: Improving the Performance of Non-Thinking Models at No Cost
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.
-
CircuitFormer: A Circuit Language Model for Analog Topology Design from Natural Language Prompt
CircuitFormer is a 511M-parameter encoder-decoder model that generates analog circuit topologies from text prompts at 100% syntactic correctness and 83% functional success using a new subcircuit-mining tokenizer that keeps vocabulary size fixed at 512.
-
RAG over Thinking Traces Can Improve Reasoning Tasks
Retrieving structured thinking traces as a corpus improves reasoning performance on AIME, LiveCodeBench, and GPQA over standard RAG or no retrieval.
-
It Takes a MAESTRO To Prune Bad Experts
Pruning MoE LLMs according to the stationary distribution of a Markov chain over (layer, expert) routing transitions retains more task performance than local importance heuristics, with up to ~3.5% relative gains over the strongest baseline at 50% compression.
-
Sparse Delta Memory: Scaling the State of Linear RNNs through Sparsity
SDM sparsifies the Gated DeltaNet update rule to enable 1000x larger recurrent memory states at iso-FLOP, improving long-context recall and short-context reasoning over GDN and matching full attention at 8B scale.
-
When Calibration Rankings Reverse: Accuracy-Controlled Evaluation for Fair Comparison of LLMs
Global calibration metrics like ECE are confounded by accuracy; the proposed ACE framework with three accuracy-controlled views shows many prior calibration advantages weaken or reverse.
-
SVR-MAD: A Bayesian-Inspired Framework for Posterior-Guided Multi-Agent Debate
SVR-MAD treats pre-debate signals as priors and debate results as evidence to build a sparser communication graph, cutting token use by up to 61% while preserving or raising accuracy over prior MAD methods.
-
SWE-Mutation: Can LLMs Generate Reliable Test Suites in Software Engineering?
SWE-Mutation benchmark shows current LLMs achieve low verification (10.20%) and detection (36.15%) rates on 2,636 mutated variants, exposing weaknesses in generating reliable test suites.
-
How Far Will They Go? Red-Teaming Online Influence with Large Language Models
An empirical red-teaming study measures political Overton Windows across more than 30 open-source LLMs from 10 families and finds left-leaning bias, inverse size correlation, regional variation, and variable jailbreak effectiveness.
-
The Illusion of Intervention: Your LLM-Simulated Experiment is an Observational Study
Interventions in LLM-simulated user experiments induce distribution shifts in latent attributes that create confounding bias, diagnosable with negative control outcomes and partially mitigated by adding setting-relevant persona details.
-
AMARIS: A Memory-Augmented Rubric Improvement System for Rubric-Based Reinforcement Learning
AMARIS augments rubric updates in RL for LLMs with a persistent memory of rollout analyses and prior edits, yielding gains such as +2.8 points on GPQA-Diamond over local-adaptive baselines.
-
Predictive Prefetching for Retrieval-Augmented Generation
Introduces predictive prefetching for RAG that anticipates retrieval needs several tokens ahead via three components, reporting up to 43.5% latency reduction and 62.4% TTFT improvement while preserving answer quality.
-
Self-Pruned Key-Value Attention: Learning When to Write by Predicting Future Utility
SP-KV trains a utility predictor jointly with the LLM to dynamically prune low-utility KV cache entries, achieving 3-10x memory reduction during generation with negligible performance loss.
-
BioTool: A Comprehensive Tool-Calling Dataset for Enhancing Biomedical Capabilities of Large Language Models
BioTool dataset enables fine-tuning a 4B-parameter LLM to outperform GPT-5.1 in biomedical tool calling while improving downstream answer quality per human experts.
-
MultiBreak: A Scalable and Diverse Multi-turn Jailbreak Benchmark for Evaluating LLM Safety
MultiBreak is a large diverse multi-turn jailbreak benchmark that achieves substantially higher attack success rates on LLMs than prior datasets and reveals topic-specific vulnerabilities in multi-turn settings.
-
Verbal-R3: Verbal Reranker as the Missing Bridge between Retrieval and Reasoning
Verbal-R3 uses a verbal reranker to generate analytic narratives that guide retrieval and reasoning in LLMs, achieving SOTA results on complex QA benchmarks.
-
LLM Output Detectability and Task Performance Can be Jointly Optimized
Fine-tuning an LLM with DPO against a detector score plus a task metric (PUPPET) improves both the detector's AUROC and downstream task scores, beating watermarked baselines on task quality in the reported benchmarks.
-
Disentangling Linguistic Relatedness from Task Alignment in Cross-Lingual Transfer
Fine-tuning LLMs on Arabic yields similar zero-shot gains on Semitic and non-Semitic languages, with chain-of-thought reasoning producing parallel benefits, indicating task alignment drives transfer more than language relatedness.
-
Polysemantic Experts, Monosemantic Paths: Routing as Control in MoEs
A parameter-free decomposition in MoE models separates routing control from content, showing that expert trajectories cluster tokens by semantic function across languages and forms, making paths rather than experts the natural unit of interpretability.
-
SPENCE: A Syntactic Probe for Detecting Contamination in NL2SQL Benchmarks
SPENCE shows older NL2SQL benchmarks like Spider have high performance sensitivity to syntactic changes, indicating likely training contamination, while newer ones like BIRD show little sensitivity and appear largely clean.
-
Exploring the Effectiveness of Using LLMs for Automated Assessment of Student Self Explanations in Programming Education
Compares LLMs against semantic similarity for binary classification of student self-explanations in programming education.
-
ReacTOD: Bounded Neuro-Symbolic Agentic NLU for Zero-Shot Dialogue State Tracking
ReacTOD introduces a bounded neuro-symbolic ReAct architecture with symbolic validation that delivers new zero-shot SOTA joint goal accuracy on MultiWOZ 2.1 and strong results on SGD.
-
NoisyCoconut: Counterfactual Consensus via Latent Space Reasoning
Injecting noise into LLM latent trajectories creates diverse reasoning paths whose agreement acts as a confidence signal for selective abstention, cutting error rates from 40-70% to under 15% on math tasks.
-
NVIDIA Nemotron 3: Efficient and Open Intelligence
NVIDIA releases the Nemotron 3 model family with hybrid Mamba-Transformer architecture, LatentMoE, NVFP4 training, MTP layers, and multi-environment RL post-training for reasoning and agentic tasks.
-
Supplement Generation Training for Enhancing Agentic Task Performance
SGT trains a lightweight model to generate task-specific supplemental text that improves performance of a larger frozen LLM on agentic tasks without modifying the large model.
- One prompt is not enough: Instruction Sensitivity Undermines Embedding Model Evaluation
- Convergent Evolution: How Different Language Models Learn Similar Number Representations