VerifySteer selectively steers hidden states at paragraph boundaries using latent correctness signals to control verifier strictness and outperform baselines on ProcessBench and Hard2Verify with lower compute.
Title resolution pending
17 Pith papers cite this work, alongside 191 external citations. Polarity classification is still indexing.
17
Pith papers citing it
191
external citations · OpenAlex
citation-role summary
background 4
citation-polarity summary
roles
background 4polarities
background 4representative citing papers
Presents a non-linear intervention framework for LLMs with a learning procedure for implicit features, validated on refusal bypass steering showing improved precision over linear baselines.
AgentFlow uses a typed graph DSL covering roles, prompts, tools, topology and protocol plus a runtime-signal feedback loop to optimize multi-agent harnesses, reaching 84.3% on TerminalBench-2 and discovering ten new zero-days in Chrome including two critical sandbox escapes.
PersonalHomeBench is a new benchmark showing that AI agents suffer systematic performance drops in personalized smart homes as task complexity rises, especially in counterfactual reasoning and partial observability.
A new parallel reasoning dataset enables LLMs to shift reasoning to non-English languages via SFT and RLVR while matching or exceeding baseline performance.
FlexMoE produces nested pruned subnetworks for MoE LLMs across budgets via channel importance ranking and discrete action learning, plus one mid-budget recovery fine-tune, retaining 99.8% performance at 50% expert parameter pruning.
Fine-tuning language encoding models on fMRI responses improves prediction performance for ECoG brain signals in frequency bands beyond fMRI resolution.
Multi-agent LLM teams outperform human teams in creativity (d=1.50) across tasks by producing more novel ideas, with distinct semantic exploration patterns predicting success for each group.
SFT on divergent branch-heavy CoT from DeepSeek-R1 yields worse generalization than convergent CoT from gpt-oss despite lower loss, but filtering frequent branches improves average performance by 3.6% on five reasoning benchmarks.
SitEmb-v1.5 uses a new training paradigm to produce context-situated embeddings for short chunks, outperforming larger models by over 10% on a curated book-plot retrieval benchmark.
RetroInfer introduces the wave index and wave buffer to realize sparse KV-cache attention for long-context LLM inference with up to 4.4X throughput gains while matching full-attention accuracy.
TRACE is a new metric for assessing LLM CoT reasoning structure via Toulmin and Flavell frameworks, showing r=0.74 correlation with accuracy on 26.3K samples and utility as an RL reward.
Combines GRPO with teacher-guided on-policy distillation and introduces LongBlocks dataset to yield more stable long-context reasoning than either method alone.
Reshaping outcome rewards, process signals, and rollout comparability in GRPO raises strict compile-and-semantic accuracy in agentic code repair from 0.385 to 0.535 under weak feedback.
MathArena is broadened into a maintained platform with new benchmarks for proofs, research questions, and formal verification, where GPT-5.5 scores 98% on 2026 USAMO and 74% on research-level tasks.
MidSteer is presented as a general optimal affine framework for directed, minimal-disturbance concept steering of generative models, with standard erasure as a special case of LEACE.
Mira-Embeddings-V1 adapts embeddings for recruitment reranking by synthesizing positive and hard-negative samples with LLMs, then applies JD-JD contrastive and JD-CV triplet training plus a BoundaryHead MLP, lifting Recall@50 from 68.89% to 77.55% and Recall@200 from 0.5969 to 0.7047.
citing papers explorer
-
The Hidden Signal of Verifier Strictness: Controlling and Improving Step-Wise Verification via Selective Latent Steering
VerifySteer selectively steers hidden states at paragraph boundaries using latent correctness signals to control verifier strictness and outperform baselines on ProcessBench and Hard2Verify with lower compute.
-
Non-linear Interventions on Large Language Models
Presents a non-linear intervention framework for LLMs with a learning procedure for implicit features, validated on refusal bypass steering showing improved precision over linear baselines.
-
Synthesizing Multi-Agent Harnesses for Vulnerability Discovery
AgentFlow uses a typed graph DSL covering roles, prompts, tools, topology and protocol plus a runtime-signal feedback loop to optimize multi-agent harnesses, reaching 84.3% on TerminalBench-2 and discovering ten new zero-days in Chrome including two critical sandbox escapes.
-
PersonalHomeBench: Evaluating Agents in Personalized Smart Homes
PersonalHomeBench is a new benchmark showing that AI agents suffer systematic performance drops in personalized smart homes as task complexity rises, especially in counterfactual reasoning and partial observability.
-
ReasonXL: Shifting LLM Reasoning Language Without Sacrificing Performance
A new parallel reasoning dataset enables LLMs to shift reasoning to non-English languages via SFT and RLVR while matching or exceeding baseline performance.
-
FlexMoE: One-for-All Nested Intra-Expert Pruning for MoE Language Models
FlexMoE produces nested pruned subnetworks for MoE LLMs across budgets via channel importance ranking and discrete action learning, plus one mid-budget recovery fine-tune, retaining 99.8% performance at 50% expert parameter pruning.
-
Fine-tuning language encoding models on slow fMRI improves prediction for fast ECoG
Fine-tuning language encoding models on fMRI responses improves prediction performance for ECoG brain signals in frequency bands beyond fMRI resolution.
-
Multi-agent AI systems outperform human teams in creativity
Multi-agent LLM teams outperform human teams in creativity (d=1.50) across tasks by producing more novel ideas, with distinct semantic exploration patterns predicting success for each group.
-
On the Role of Reasoning Patterns in the Generalization Discrepancy of Long Chain-of-Thought Supervised Fine-Tuning
SFT on divergent branch-heavy CoT from DeepSeek-R1 yields worse generalization than convergent CoT from gpt-oss despite lower loss, but filtering frequent branches improves average performance by 3.6% on five reasoning benchmarks.
-
SitEmb-v1.5: Improved Context-Aware Dense Retrieval for Semantic Association and Long Story Comprehension
SitEmb-v1.5 uses a new training paradigm to produce context-situated embeddings for short chunks, outperforming larger models by over 10% on a curated book-plot retrieval benchmark.
-
RetroInfer: A Vector Storage Engine for Scalable Long-Context LLM Inference
RetroInfer introduces the wave index and wave buffer to realize sparse KV-cache attention for long-context LLM inference with up to 4.4X throughput gains while matching full-attention accuracy.
-
TRACE: Toulmin-based Reasoning Assessment through Constructive Elements for LLM CoT Evaluation
TRACE is a new metric for assessing LLM CoT reasoning structure via Toulmin and Flavell frameworks, showing r=0.74 correlation with accuracy on 26.3K samples and utility as an RL reward.
-
A Recipe for Long-Context Reasoning in Large Language Models via On-Policy Optimization and Distillation
Combines GRPO with teacher-guided on-policy distillation and introduces LongBlocks dataset to yield more stable long-context reasoning than either method alone.
-
Signal Reshaping for GRPO in Weak-Feedback Agentic Code Repair
Reshaping outcome rewards, process signals, and rollout comparability in GRPO raises strict compile-and-semantic accuracy in agentic code repair from 0.385 to 0.535 under weak feedback.
-
Beyond Benchmarks: MathArena as an Evaluation Platform for Mathematics with LLMs
MathArena is broadened into a maintained platform with new benchmarks for proofs, research questions, and formal verification, where GPT-5.5 scores 98% on 2026 USAMO and 74% on research-level tasks.
-
MidSteer: Optimal Affine Framework for Steering Generative Models
MidSteer is presented as a general optimal affine framework for directed, minimal-disturbance concept steering of generative models, with standard erasure as a special case of LEACE.
-
Mira-Embeddings-V1: Domain-Adapted Semantic Reranking for Recruitment via LLM-Synthesized Data
Mira-Embeddings-V1 adapts embeddings for recruitment reranking by synthesizing positive and hard-negative samples with LLMs, then applies JD-JD contrastive and JD-CV triplet training plus a BoundaryHead MLP, lifting Recall@50 from 68.89% to 77.55% and Recall@200 from 0.5969 to 0.7047.