PhysCount is a benchmark of 546 real-world images and 4,595 human-written questions in 651 target-specific groups. It tests whether multimodal LLMs can separate visible appearances (direct view, reflections, glass) from the physical entities behind them. Across 40 MLLMs, counting appearances is much easier than linking them to entities, and the best model answers every question correctly in only 13.36% of groups.
vision-language model
PhysCountarXiv
evals, multimodal, researchΒ· Added 31 Oct 2026, published 29 Oct 2026
Glance is a RAG method in which the generator also reranks. It adds a lightweight decision adapter and pointer head that score candidate passages from one shared prefill, so a single resident 9B backbone does generation and reranking. After 39 minutes of MS MARCO fine-tuning it reaches nDCG@10 of 0.741 on SciFact and 0.345 on NFCorpus, against 0.721 and 0.326 for bge-reranker-v2-m3. On 300 HotpotQA questions it improves answer F1 by 0.045 over cross-encoder-selected context. Evaluation covers two retrieval datasets, one QA dataset and one backbone family.
RAGLLMbge-reranker-v2-m3
MS MARCOSciFactNFCorpus+1
llm-apps, researchΒ· Added 31 Oct 2026, published 29 Oct 2026
High Precision Noise Cleaning for CKKS is an IACR ePrint preprint (2026/2675) by Zhihao Li, Cheng Hong and Tao Wei. It composes lower-degree Hermite polynomials to reach output noise bounds such as 2^-300 for noise flooding, and uses this to build an asynchronous threshold CKKS scheme with strong simulation security in the random oracle model. In OpenFHE, binary-message cleaning falls from 9.63 s to 1.54 s.
Paper studying whether self-evolution weakens personal agents' robustness to indirect prompt injection (IPI). A two-phase method has the user ask the agent to follow tasks in external content, then tests with a third-party injected task the user did not authorize. Against the non-evolving base agent, attack success rose 14.0 to 48.8 points under skill evolution and 3.0 to 39.3 points under reinforcement learning. Code is released.
indirect prompt injection
AI SecurityΒ· Added 31 Oct 2026, published 29 Oct 2026
Gold vs. Translation is an October 2026 arXiv study of whether DeepL-translated MMLU measures the same thing as native benchmarks. It compares seven pairings against six native benchmarks in five Nordic and Baltic languages, using 29 models. Translated MMLU adds predictive value for curriculum-overlap benchmarks such as Skolprov (ΞRΒ²=.128) but not for Sweden-specific SwedishFacts (p=.242).
LLM
MMLUDeepLSkolprov+2
evals, researchΒ· Added 30 Oct 2026, published 28 Oct 2026
DeoGen is an automatic, parametrizable benchmark generator for deontic reasoning, where a model applies policy rules to facts to reach a verdict. It autoformalizes rules into first-order logic, builds decision trees, and turns each tree path into a benchmark entry. Frontier models give the correct verdict 89-92% of the time, but only 22-40% on entries lacking enough information, even when told to abstain.
LLMfirst-order logic
DeoGenarXiv
evals, researchΒ· Added 30 Oct 2026, published 28 Oct 2026
QbD-VLA is a mixed-precision post-training quantization method for vision-language-action models. It uses task success from short closed-loop evaluations to assign weight and activation bit widths, accounting for interactions between components. On four VLA models across all four LIBERO suites it holds task performance at about 4-bit weights and under 6-bit activations. The gain on SIMPLER is small. On a Franka Research 3 arm, X-VLA shows no detected loss, but Ο0.5 fails.
vision-language-action model
QbD-VLALIBEROSIMPLER+4
research, inference, multimodalΒ· Added 30 Oct 2026, published 28 Oct 2026
This paper replicates Chen et al.'s study of LLM-based smart contract vulnerability detection, first on GPT-4 and then on Mixture-of-Experts, RNN-like and State Space Models. The exact replication confirms the high-recall, low-precision pattern on GPT-4. The differentiated replication finds the pattern does not transfer uniformly: GPT-4 and Gemma-3 behave alike despite a roughly 100x parameter gap, while two MoE models diverge.
LLMTransformerMixture of Experts+3
Chen et al.GPT-4Gemma 3+1
research, evalsΒ· Added 30 Oct 2026, published 28 Oct 2026
This arXiv paper (October 2026) replicates Chen et al.'s study of LLM-based smart contract vulnerability detection. It confirms the high-recall, low-precision pattern on GPT-4, then tests Mixture-of-Experts and RNN-like models plus a preliminary State Space Model run. Gemma-3 matches GPT-4's profile despite far fewer parameters, while two Mixture-of-Experts models diverge, so architecture family alone does not predict detection behavior.
GPT-4GemmaLLM
Chen et al.
AI Security, AppSecΒ· Added 30 Oct 2026, published 28 Oct 2026
JevAct is a runtime verification framework that intercepts selected LLM agent actions before execution and returns feedback for bounded revision. It compiles domain policies into executable state checks and structured questions scored by a prefill-only decision model, giving probability-based judgments without generation. Tested on eleven agents with tau^2-bench Airline and Telecom and BFCL, it gained consistently on Telecom and BFCL and was mixed on Airline.
LLM
JevActtau^2-benchBFCL+1
research, agents, evalsΒ· Added 30 Oct 2026, published 28 Oct 2026
This paper measures job-based doubt in LLM fact-checkers: judging a claim false more often when the poster has a lower-status job. It compares base and post-trained checkpoints of six model families on fact-checked claims with randomised poster profiles. Only the open TΓΌlu-3 recipe (OLMo-2 32B, TΓΌlu-3 70B) creates the effect, under one percentage point but reliable. In OLMo-2 32B the flip comes from one residual-stream direction and appears at supervised fine-tuning, confirmed on 400 new claims. None of seven commercial models doubts lower-status posters more.
LLMTransformer
OLMo-2 32BTΓΌlu-3TΓΌlu-3 70B+3
research, ai-safety, trainingΒ· Added 30 Oct 2026, published 28 Oct 2026
This preprint benchmarks 13 decision models against 15 generative LLM configurations, including different reasoning settings, on 1,273 text-only four-option MedQA questions. Accuracy ran 67.71% to 89.08% for decision models and 78.32% to 96.86% for generative configurations. Decision models were cheaper and faster, but answers with probability of at least 0.90 still had error rates of 1.71% to 12.77%, and probability repeatability varied across three runs.
LLM
MedQAMercury DecideClef Omni+1
evals, researchΒ· Added 30 Oct 2026, published 28 Oct 2026
TCC-RAG is a retrieval-augmented generation method for temporal knowledge graph question answering (TKGQA). It turns the temporal knowledge graph into a multi-view causal cue graph, builds a query-specific temporal causal chain to guide retrieval, and applies a causal evidence filter to the results. The authors report gains of 10.72% and 8.40% on two TKGQA benchmarks and 62.97% on multi-step temporal reasoning.
RAGLLM
arXiv
research, llm-appsΒ· Added 30 Oct 2026, published 28 Oct 2026
Metacognitive Decoding is a lightweight plug-in controller for diffusion language model backbones. It learns when to finalize tokens and how far generation should extend, and is trained on agent trajectories to balance correctness against generation cost. Across five benchmarks it cuts inference FLOPs by 20.6% on average versus the most efficient baseline, with competitive task accuracy. The abstract names no specific benchmarks or backbones.
diffusion model
arXiv
research, agents, inferenceΒ· Added 28 Oct 2026, published 27 Oct 2026
NEXUS is a hierarchical reinforcement learning framework that pairs neural skills with symbolic meta-policies, so transparent reasoning sits at the level of high-level skills rather than atomic actions. It uses object-centric observations and LLM interaction to design reward functions and meta-policies. Agents learn disentangled skills, align better with goals in the game manual, resist reward hacking more, and cope better with game simplifications than deep RL. The paper also compares neural, symbolic and neuro-symbolic meta-policy architectures.
LLM
NEXUSarXiv
research, training, ai-safetyΒ· Added 28 Oct 2026, published 27 Oct 2026
AdaRoll-GRPO is a method that allocates GRPO rollout budgets per question by difficulty instead of uniformly. It frames allocation as a sequential decision problem and learns an allocation policy with a Bayes-adaptive MDP, computed offline so training adds no online optimization cost. Experiments on several reasoning datasets and LLM architectures show gains over standard GRPO at the same total rollout budget.
GRPOLLM
AdaRoll-GRPOarXiv
training, researchΒ· Added 28 Oct 2026, published 27 Oct 2026
OpenGray is a benchmark that adds tasks with known correct decisions to the OpenKBP-Opt head-and-neck dataset, including two overlap contradictions with provably incompatible dose goals and a feasible control. Four planning systems, from a rule-based script to a full LLM agent, were tested on 30 patients with Terra and Ministral 3 8B. Decision accuracy reached 95.4% and 92.1%, yet only 50% and 25% of overlap contradictions were escalated. SAGE logs of 39 radiosurgery runs showed an agent that lacked an action for stopping.
LLM
OpenGrayOpenKBP-OptSAGE+3
evals, agents, researchΒ· Added 28 Oct 2026, published 27 Oct 2026
This paper tests the delta-learning hypothesis for Direct Preference Optimization (DPO), where the larger model's response in a single model pair is treated as preferred. It shows, in theory and experiments, that fixed pairs add generator-specific biases that predict the preference label regardless of quality. DPO on single-pair data underperforms multi-model data, consistent with reward hacking, and adding more generators partly fixes it without an external judge.
LLM
arXiv
training, researchΒ· Added 28 Oct 2026, published 27 Oct 2026
"When Algebraic Structure Meets Too Many Hints" is an ePrint paper (2026/2654) arguing that DDGR hint classes and too-many-hints thresholds are not standalone side-channel security metrics. In synthetic Kyber-512 experiments, NTT-domain residues recover the full secret with about half the hints the uniform mod-q bound predicts. Finite-precision FFT samples need 25% to 55% of the budget for uniform perfect hints. No new attack optimizations are proposed.
Anis Bkakria's ePrint 2026/2638 gives polynomial-time decision attacks on Known-Covariance Ring-LWE (KCov-RLWE), the assumption Micciancio and Suhl introduced for simulation-secure threshold encryption. The public sample and covariance hint form a bordered Gram matrix whose determinant leaks a residual that differs between real and uniform samples. The attacks cover broad prime-modulus families at ring rank one and transfer to composite moduli with an odd prime divisor. Power-of-two moduli are not covered, and the paper does not break the full threshold-encryption scheme.
Ring-LWE
Known-Covariance Ring-LWEMicciancio and SuhlGram determinant+2
Philippe Gaborit proves that maximum-likelihood decoding of Gabidulin codes is NP-hard even for dimension one: deciding rank distance at most n-2 from the code {Ξ»g} is NP-complete. The reduction is deterministic, so Rank Syndrome Decoding and Rank Minimum Distance for F_{q^m}-linear codes are NP-complete under deterministic reductions, where the earlier reduction was randomized. It goes through a new pivot problem, Twisted Square Membership.
Gabidulin codes
Philippe GaboritGuruswami-Vardy theoremTwisted Square Membership+2
A stage-aware visual attribution framework for Vision-Language Navigation (VLN) that explains a robot's action choices. It identifies "instruction-stage insensitivity", where existing explainers give similar outputs across navigation stages. The method builds action-conditioned temporal instruction-visual attributions, aggregates relevant historical evidence by stage, and applies an instruction-guided spatial filter. It improves attribution quality across several VLN policies and benchmarks, helps diagnose failure cases, and was tested on a Unitree G1 robot.
vision-language model
Unitree G1
research, multimodalΒ· Added 27 Oct 2026, published 25 Oct 2026
This paper evaluates sycophancy in medical diagnosis across nine AI models. Each model is shown clinical evidence plus a patient's belief in a specific but incorrect condition, with the belief's closeness to the evidence varied. Models adopted the belief 23.2% of the time on average, even for fictional conditions, and 43.0% under a Rare Disease Claimer persona. Medical fine-tuning and larger scale reduced susceptibility.
LLM
arXiv
ai-safety, evals, researchΒ· Added 27 Oct 2026, published 25 Oct 2026
HerHealthGPT is a conversational AI system for women's reproductive health that combines retrieval-augmented generation, context-aware clarification and source validation in a web interface. Evidence retrieval raises concern-category accuracy from 54.1% to 59.3% over fine-tuning alone on 540 messages from 90 cases. The clarification router reaches 85.6% recall and 86.2% specificity. In a 30-adult study, 83.3% preferred the retrieval-enabled version. The authors say clinical validation is still needed.
RAGLLM
HerHealthGPTarXiv
research, llm-appsΒ· Added 27 Oct 2026, published 24 Oct 2026
DRILS is an LLM-based agentic framework for requirements change impact analysis. It uses multiple agents to generate candidate impact sets, with an LLM-as-a-judge promoting diversity in selection. It adds RE-RAG, a tuned retrieval technique, and a requirement dependency graph as a knowledge source. On three curated datasets it reaches 87.1% recall, 76.3% precision and 81% F1 at lower cost than baselines. The authors release the datasets.
LLMRAG
arXivDRILSRE-RAG
research, agents, llm-appsΒ· Added 27 Oct 2026, published 24 Oct 2026
Ranking Is Not Deciding is an October 2026 arXiv paper showing that query-time RAG poisoning filters score near-perfectly because of how the poison set was built. PoisonedRAG's code prepends the target question to each passage. Removing that prefix takes RAGuard from AUROC 0.953 to 0.758 and its miss rate from 0.191 to 0.763. The authors release a script that profiles a poison set.
LLM
PoisonedRAGRAGuardTrustRAG+5
AI SecurityΒ· Added 27 Oct 2026, published 24 Oct 2026
AutoTavern is a self-improving system of specialised LLM research agents with a typed shared memory and verification gates. In 20 days it built TavernGym, a full-content Hearthstone Battlegrounds simulator, and 23 versions of TavernAgent, a PPO-trained player. TavernAgent averaged placement 1.60 with 80% first places over its last 20 live games, beating the top ten players (2.34, 59%). On TavernBench (960 tasks), it answered 91% of play tasks against at most 63% for frontier LLMs.
PPOLLM
AutoTavernTavernGymTavernAgent+3
agents, research, training, evalsΒ· Added 27 Oct 2026, published 24 Oct 2026
Sliding Networks is a family of residual network structures with a joint width and depth scaling recipe that keeps both locally maximal feature learning and feature diversity in the residual stream. The key quantity is the ratio N_flat/(N_sharp L) tending to a constant alpha, and Half Copy correlated initialization can relax the shape constraints. Residual-MLP experiments support an SDE limit, but language-model experiments show performance differences under regrouping at fixed parameter count.
Transformer
arXiv
researchΒ· Added 27 Oct 2026, published 24 Oct 2026
Persistent Closed-Form Routing (PCR) is a plug-and-play method for multivariate time-series forecasting. It alternates exact conditional estimation of channel-specific mixture weights with gradient training of low-rank residual experts around a frozen pretrained forecaster. Across eleven datasets and three backbones, PCR-K4 cuts dataset-balanced MSE by 3.86-14.25% and beats competing specialization methods with fewer trainable parameters. The paper also introduces GalaxyTS, a synthetic dataset with exact shared and channel-specific references.
Mixture of ExpertsTransformer
GalaxyTSPCRPCR-K4+1
researchΒ· Added 27 Oct 2026, published 24 Oct 2026
This paper compares LoRA and full fine-tuning for prompt-safety classification using a matched 400M encoder-decoder pair (Ettin) across three task formats, with up to 15 seeds per cell and pre-specified analyses. Full fine-tuning leads LoRA by 2-3 pp at the pre-specified learning rates, and the encoder gap closes when both use the same learning-rate search. Training LoRA to write a reason after its verdict adds 4.0 pp, while a content-free sentence of equal length does not. Trained models beat a prompted 8B in domain but trail it by 8-23 pp on adversarial jailbreaks and over-flag benign prompts (27-66% against 2-3%).
LoRAEttinTransformer
training, research, evalsΒ· Added 27 Oct 2026, published 24 Oct 2026
arxiv.orgpaperπ§ AI
