Persona vectors form within the first 0.22% of LLM pretraining and remain effective for steering post-trained models, with continued refinement and transfer to other models.
hub Mixed citations
Deepseek-v4: Towards highly efficient million-token context intelligence
Mixed citation behavior. Most common role is background (62%).
26
Pith papers citing it
Background
62% of classified citations
hub tools
citation-role summary
background 8
baseline 5
citation-polarity summary
years
2026 26representative citing papers
A new native-runtime benchmark reveals that current frontier AI agents succeed on at most 62 percent of realistic long-horizon CLI tasks.
MathConstraint generates scalable, automatically verifiable combinatorial problems where LLMs achieve 18.5-66.9% accuracy without tools but roughly double that with solver access.
Introduces a cost-aware paired protocol with six outcome groups and applies it to Dynamic-SAGE versus SAGE, reporting 7.5-point accuracy gain, 28% fewer tool calls, but 34% higher token use.
LLM-generated research ideas cluster more around bridge-like opportunities and synthesis methods than the broader distribution seen in human papers.
WebGameBench is a new benchmark that evaluates coding agents on building browser-native games from frozen specifications, with runtime browser evaluation showing best agents reach 76.9% usable rate but only 20.2% excellent rate.
On power-law covariance least squares problems, SignSVD (Muon) and SignSGD (Adam proxy) show three phases of relative performance depending on data exponent α and target exponent β.
LoopUS converts pretrained LLMs into looped latent refinement models via block decomposition, selective gating, random deep supervision, and confidence-based early exiting to improve reasoning performance.
A new Intent Fidelity Score and refinement loop verify that LLM-generated simulation code matches the intended PDEs, improving performance on a 220-case benchmark where execution alone fails to ensure correctness.
AgentForesight introduces an online auditor model that predicts decisive errors in multi-agent trajectories at the earliest step using a coarse-to-fine reinforcement learning recipe on a new curated dataset AFTraj-2K.
Rubric-based on-policy distillation allows training student models using only teacher responses by generating scoring rubrics from contrasts and using them for on-policy optimization, achieving superior performance and up to 10x better sample efficiency than logit-based approaches.
Muon with Nesterov momentum and inexact polar decomposition achieves optimal convergence rates of O(ε^(-(3α-2)/(α-1))) under heavy-tailed noise for ε-stationary points in non-convex settings.
Joint Consistency casts test-time aggregation as Ising-type energy minimization with pairwise LLM-judge interactions, subsuming voting methods and outperforming baselines across reasoning tasks.
HarnessX assembles and evolves agent harnesses via substitution algebra and AEGIS trace analysis, reporting +14.5% average gains (up to +44%) on five benchmarks.
OpenSkillEval dynamically builds task instances across five application domains to evaluate 30 open skills with over 600 tests, finding that skill use depends heavily on model and framework and that many popular skills do not beat base agents.
Vision-OPD transfers an MLLM's privileged regional perception to its full-image policy through on-policy token-level self-distillation, yielding competitive results on fine-grained visual benchmarks.
RESD turns failure trajectories into token-level supervision via retrospective reflections and a persistent global playbook, enabling faster improvement than standard self-distillation or GRPO with only one rollout per prompt.
DeltaRubric decomposes multimodal preference evaluation into self-generated planning and verification steps within a single model, producing large accuracy improvements on VL-RewardBench via multi-role reinforcement learning.
VeriContest supplies 946 problems with specs, code, proofs, and tests to benchmark verifiable code generation in Rust/Verus, showing models reach 92% on code but only 5% end-to-end on full verifiable synthesis.
LogiHard hardens reasoning benchmarks by transforming 0-order selection into 2-order judgment, causing 31-56% accuracy drops in 12 frontier LLMs and a 47% drop on zero-shot MMLU, revealing a combinatorial reasoning gap rather than knowledge deficits.
Behavior Cue Reasoning trains LLMs to emit special tokens before behaviors, enabling monitors to cut up to 50% wasted reasoning tokens and recover safe actions from 80% of unsafe traces, more than doubling success rates with no performance cost.
CoPD integrates multiple expert capabilities by running parallel RLVR training with bidirectional online policy distillation among experts, outperforming mixed RLVR and sequential OPD while surpassing domain-specific experts on text-image-video reasoning.
TaskGround introduces a Ground-Infer-Execute framework for full-scene household reasoning that improves success rates on the FullHome benchmark and enables compact models to match larger ones at up to 18x lower token cost.
CompactAttention accelerates chunked-prefill attention via Block-Union KV Selection, delivering up to 2.72x speedup at 128K context on LLaMA-3.1-8B while matching dense accuracy on RULER.
citing papers explorer
-
WebGameBench: Requirement-to-Application Evaluation for Coding Agents via Browser-Native Games
WebGameBench is a new benchmark that evaluates coding agents on building browser-native games from frozen specifications, with runtime browser evaluation showing best agents reach 76.9% usable rate but only 20.2% excellent rate.
-
Joint Consistency: A Unified Test-Time Aggregation Framework via Energy Minimization
Joint Consistency casts test-time aggregation as Ising-type energy minimization with pairwise LLM-judge interactions, subsuming voting methods and outperforming baselines across reasoning tasks.
-
HarnessX: A Composable, Adaptive, and Evolvable Agent Harness Foundry
HarnessX assembles and evolves agent harnesses via substitution algebra and AEGIS trace analysis, reporting +14.5% average gains (up to +44%) on five benchmarks.
-
Behavior Cue Reasoning: Monitorable Reasoning Improves Efficiency and Safety through Oversight
Behavior Cue Reasoning trains LLMs to emit special tokens before behaviors, enabling monitors to cut up to 50% wasted reasoning tokens and recover safe actions from 80% of unsafe traces, more than doubling success rates with no performance cost.
-
TaskGround: Structured Executable Task Inference for Full-Scene Household Reasoning
TaskGround introduces a Ground-Infer-Execute framework for full-scene household reasoning that improves success rates on the FullHome benchmark and enables compact models to match larger ones at up to 18x lower token cost.
-
The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixes
On-policy self-distillation fails for instance-specific privileged information because the student learns an aggregated PI-free policy, while on-policy distillation is sensitive to teacher choice and loss formulation, with stop-gradient and stabilized methods as mitigations.