Shows that fully quantizing all 496 linear layers—including the recurrent Gated DeltaNet—of a hybrid 27B LLM to 4-bit NVFP4 W4A4 preserves benchmark accuracy while reducing model size to 17.5 GiB and improving prefill speed; includes a calibrated NVFP4 checkpoint.
Compresses KV cache for long-chain reasoning by keeping prompt tokens and evicting remaining entries uniformly at random per attention head; across four models and six reasoning tasks it matches the strongest prior evictor while delivering 32–43% higher vLLM throughput. Relies on prompt protection and redundancy across heads/text to retain reasoning traces; suitable when static memory budgets and higher serving throughput are priorities.
AI Agent Papers2026
Sihan Ge, Yichen Lin +4Affiliation: Cardinal Operations, Affiliation: Shanghai Jiao Tong University, Shanghai, China
[email protected] {linyichen, chenyuzhou, linjianghao, taoyao, ddge}@sjtu.edu.cn
* Equal contribution † Corresponding authors
Studies when and how an assistant should ask clarification questions before converting natural-language problem descriptions into optimization models. Introduces OR-Clarify, a benchmark for pre-formulation clarification, and InterOPT, a two-stage framework that diagnoses formulation-critical gaps and guides targeted questioning and stopping decisions.
Provides a diagnostic benchmark of 4,084 multiple-choice questions derived from longitudinal wearable time series, blood biomarkers, and demographics of 200 users to evaluate LLM reasoning on real-world wearable data. Uses 16 question types across complementary axes and a dual-grounding framework combining literature findings with population statistics.
AI Agent Papers2026
Transforms source code into verifiable, reusable agent skills by extracting atomic operations, workflows, and recurring patterns and validating them via source-body-blind reconstruction. Produces CodeSkillBank (1,006,822 accepted records from 19,769 GitHub repos) and yields ~11.7% average downstream improvement.
AI Agent Papers2026
Synthesizes a deployment-specific safety harness (natural-language policy plus executable logic) for a frozen LLM in a target domain, using model behavior, domain specs, and adversarial review to optimize safety–utility trade-offs and transfer across agent benchmarks.
Trains end-to-end driving without human trajectory supervision by decoupling perception and action: DriveVFM distills multiple frozen vision foundation models into a single camera backbone, and DriveRL trains a privileged closed-loop RL teacher whose rollouts supervise a camera-only planner, yielding state-of-the-art closed-loop benchmark results.
AI Agent Papers2026
Defines the Discovery Certification Protocol (DCP) to audit claims of discovery by AI research agents, converting claims into executable recovery and feedback tests. Specifies multi-gate certification, Core control requirements, and a deterministic offline verifier; validated in two controlled audits.
AI Model2026
Provides per-tensor mixed-precision GGUF quantizations of the Qwen3.8-Flash-Next MoE model using GSQ (Gumbel-Softmax Quantization) to produce tensor-level quantized variants and RCO (Riemannian Constrained Optimization) to allocate quant types under exact size budgets; includes a BF16 vision projector and three size/bit-width operating points.
Investigates long-horizon memorization in language models by evaluating how complementary continual-learning mechanisms (data, function, weight anchors and low-rank allocation/LoRA) compose to reduce catastrophic forgetting across 100 sequential tasks; best composition raises final retention from 1.2% to 34.9%.
AI Agent Papers2026
Yuqiao Tan, Shizhu He +2The Key Laboratory of Cognitive Intelligence, Institute of Automation, CAS, School of Artificial Intelligence, University of Chinese Academy of Sciences
Evaluates whether AI agents can autonomously use Sparse Autoencoders (SAEs) to discover interpretable features and causally steer model behavior. Introduces a 20-task benchmark that has agents design contrastive probes to search a 131K+ feature dictionary in Gemma-2-9B-IT and evaluates activation rank, selectivity, and causal steering against expert references.
Evaluates a weak teacher's RL-induced policy shift on the student's own rollouts and amplifies verifier-supported updates so stronger models can learn from weaker supervisors and surpass them. It rescales only verifier-supported policy-gradient components to preserve optimization fixed points while accelerating learning, reducing student updates versus standard RL or distillation.