SourceJul 2, 20262026-07-024 minVerified
Bug Reproduction Tests as Signals for Code Agents
How code agents can use bug reproduction tests as diagnostic signals during patch generation, not just post-hoc checks.
SourceJul 2, 20262026-07-024 minVerified
Interpreting RAG Retrieval With Sparse Autoencoder Features
Explores using sparse autoencoders to disentangle dense RAG embeddings for interpretable retrieval analysis and steering.
SourceJul 2, 20262026-07-024 minVerified
Latent Space Control for Trustworthy LLM Behavior
From steering vectors to model calibrators, this paper frames latent-space intervention as a path to better LLM control and trust.
SourceJul 2, 20262026-07-023 minVerified
On-Device AI Security Across App, Model, and OS
A look at the main security risks in mobile on-device AI, focusing on attack surfaces across apps, models, and OS.
CommunityJul 2, 20262026-07-025 minVerified
Public AI Infrastructure: Distributed Access or Concentrated Scale
Examines distributed vs. concentrated public AI compute strategies and what they mean for sovereign AI capacity.
SourceJul 1, 20262026-07-013 minVerified
Why Generator Evaluator Consistency Matters In LLM Self-Review
Why LLM self-review should be judged by generator-evaluator consistency, not accuracy alone, in agent workflows.
CommunityJun 29, 20262026-06-295 minVerified
Model Distillation, API Control, and Sovereign AI Risks
How model distillation expands from efficiency to API cost, competitive training, and control over data and compute.
SourceJun 29, 20262026-06-294 minVerified
Translating Medical AI Explanations Into Clinical Workflow
How a speech-based cognitive impairment framework turns SHAP and linguistic features into clinical explanations for usability.
SourceJun 29, 20262026-06-294 minVerified
What Should LLM Unlearning Actually Remove Precisely
A position paper argues LLM unlearning should mean dataset-defined deletion, not output suppression or behavior editing.
SourceJun 28, 20262026-06-284 minVerified
Enforcing Agent Policies Beyond Prompt-Based Safety Guards
How formalized policies can deterministically govern agent tool calls beyond probabilistic prompt steering and filters.
SourceJun 28, 20262026-06-284 minVerified
Why Benchmarks Miss Much of LLM Performance
How single-run LLM benchmarks can miss usable performance, and why model choice, retries, and cost matter.
SourceJun 27, 20262026-06-274 minVerified
Why Agent Configs Need Deterministic Control Planes
Why reused coding agent config files can become an unmanaged control layer with security and operational risks.
SourceJun 26, 20262026-06-264 minVerified
Agent-Driven Iteration Loops for Industrial Recommender Systems
A look at AgentX and the shift from model changes to automating hypothesis, code, experiment, and analysis loops.
SourceJun 26, 20262026-06-263 minVerified
Emotion Vectors in Open LLMs and Behavior Control
Examines whether emotion vectors in open-weight LLMs are internal representations or merely correlated signals for behavior.
SourceJun 26, 20262026-06-264 minVerified
HiLSVA Reframes Scientific Visualization Agent Control and Oversight
HiLSVA emphasizes plan-first workflows, human oversight, and provenance over full autonomy in scientific visualization agents.
CommunityJun 26, 20262026-06-263 minVerified
How Generative AI Makes Money And Why Profitability Debates Persist
A look at how generative AI earns revenue, why infrastructure costs loom large, and how investment and cloud deals shape profitability.
SourceJun 26, 20262026-06-265 minVerified
KARLA Rethinks Retrieval During Token Generation for LLMs
KARLA explores retrieving facts during token generation, reframing RAG tradeoffs around noise, latency, cost, and attribution.
SourceJun 26, 20262026-06-264 minVerified
Temporal Validity Challenges in RAG and Evolving Knowledge
How RAG mixes past and current facts, causing stale-fact errors, and why temporal validity matters in retrieval.
CommunityJun 25, 20262026-06-254 minVerified
Balancing AI Benefits and Existential Risks Economically
Why AI's growth benefits and existential risks should be compared within one economic framework, not separate debates.
SourceJun 25, 20262026-06-254 minVerified
Evaluating VLM Visual Search Beyond Accuracy and Tokens
A framework for evaluating VLM visual search with classic human tasks, using token length and search cost beyond accuracy.
SourceJun 25, 20262026-06-254 minVerified
Grounded LLM Workflows for Inherited Disease Diagnosis Ranking
DeepBD highlights grounded LLM workflows for inherited disease diagnosis, emphasizing traceable evidence and recall gains.
SourceJun 25, 20262026-06-254 minVerified
How LLMs Fail Plausibly on Research Math Problems
A look at four plausible LLM failure modes in research-level math and why verification design matters beyond accuracy.
SourceJun 25, 20262026-06-253 minVerified
Modeling LLM Verifier Loops With Convergence Guarantees
A framework modeling LLM-verifier loops as a four-stage absorbing Markov chain to analyze convergence and failure points.
SourceJun 25, 20262026-06-255 minVerified
Rethinking Agent Safety Beyond Model Internal Guardrails
Why agent safety must shift from internal prompts and filters to external runtime permission enforcement.