All articles
39422 articles
EnvOpt: A New Direction for Agent Environment Optimization, Cutting Costs by 72.7%
EnvOpt proposes a third direction for agent optimization: automated optimization of static environments. On AutomationBench, it cuts costs by 72.7% and improves pass rate by 14.34%, transfers across models and harnesses, and requires no graders or reference answers.
Sparse Autoencoders Reveal How Large Language Models Represent Grammatical Number
Research uses sparse autoencoders to analyze models like Pythia-70M, finding that plural noun and verb features causally control grammatical number output and revealing asymmetric singular-plural representations, offering a new method for explaining LLM grammatical mechanisms.
HyFlash: A New Prefix-Memory Paradigm for On-Device Speculative Decoding
HyFlash introduces a prefix-memory prior separating long-range context modeling from fine-grained draft-position interaction, achieving higher acceptance rates for on-device speculative decoding under limited memory with dramatically reduced KV storage and notable gains in shallow-draft and long-context scenarios.
D²-OPD: A New Method for Online Distillation that Decouples Direction and Distance
D²-OPD proposes a new method that decouples direction and distance in extrapolative distillation, using a per-prefix hard KL constraint to control target distance and adaptively combining teacher shift and initial gap to determine direction, achieving the best results across all three settings on math reasoning benchmarks, surpassing the strongest baseline by up to 9.45 points.
How Do AI Agents Waste Your Money? The Systemic Structure of Token Waste
A study proposes a unified framework to measure AI agent token waste via LLM-as-a-Judge and intervention validation, revealing system-level structure.
Archetypal Contexts in Language Model Neurons: An Architecture for Intelligence?
New research finds separable archetype, topic, and language neurons in the Qwen3.8-27B language model, with a spontaneously emerging sandwiched structure; archetypes may be the core architecture of intelligence and analogical reasoning. Code and data are open-sourced and reproducible.
Recursive Sparse Attention RecSA: Million-Token Context Extrapolation at Logarithmic Cost
RecSA recursive sparse attention uses a tree-structured KV cache for O(log L) selection cost, extrapolating from 32K training to 2M, scoring 60.5 on RULER at 1M and 3.2× faster prefill on a single H200.
SCF: A New Method for Bounded Evidence Aggregation in Language Model User Inference
SCF (Source-balanced Capped Fisher) is a bounded evidence aggregation method for language model user inference that achieves the highest AUROC across multiple datasets through capped document contributions and calibrated decision rules, with finite-sample Type-I error control.
SAFE-Teen Study: Chatbot Safety Failure Rate Reaches 43% in Teen Mental Health Conversations
SAFE-Teen study evaluated 18 chatbots in teen mental health conversations, finding a 43.3% standard failure rate without teen-aware prompts, 24.9% with prompts, and a 10–12 percentage point increase in domain-specific failures when teens push back.
Safety Vulnerabilities in Vision-Language Models: Compositional Harm and Un-alignment Auditing
New research reveals vision-language models' fragile compositional safety: high benchmark scores don't mean robustness, and un-alignment method CMaG cuts safety scores by 65 points while keeping outputs coherent, calling for internal auditing.
Anthropic Merges 30,000 Code Changes a Day — So Why Are 98% of Open-Source Issues Unclaimed?
Anthropic merges 30,000+ code changes daily, yet over 98% of issues in its public Claude Code repo are unassigned, including privacy and security bugs. Developers ask: why not use Claude for issue triage?
Indirect Prompt Injection Attacks: The Hidden Risk to AI Agents and Defense Strategies
Indirect prompt injection hides malicious instructions in content agents retrieve—documents, emails—bypassing traditional defenses to cause data exfiltration and unauthorized actions. This article explains the attack mechanism, why defenses fail, and production defense strategies.
Australian Universities Cultivate AI-N…
Australian Universities Cultivate AI-Native Graduates: ChatGPT Enters the Classroom
Australian universities are expanding access to AI tools like ChatGPT and integrating AI into teaching and course design to cultivate graduates ready for AI-enabled workplaces. This article analyzes the background of this shift, how it is implemented in teaching, and its reference significance for AI literacy education in higher education.
Claude Adds Personal Finance Analysis Tool: Privacy vs. Convenience
Claude launches Personal Finance for analyzing spending data. Reddit debates privacy: must financial data go to the cloud? Local DIY vs. cloud convenience is key.
ChatGPT GIF Feature Falls Short: User Complaints and AI Video Generation Pain Points
Reddit users criticize ChatGPT's new GIF feature: no seamless loops, garbled limbs, flickering frames, and a growing gap with voice models like Sesame. This article analyzes the technical pain points of AI video generation and user expectations.
Is the AI Trade Dead? What Reddit's Market Pullback Debate Really Reveals
A Reddit post asks whether the AI trade is dead. This article unpacks the market logic behind the emotional question, separating industry trends from trade structure and arguing that a pullback is not a disproof of the narrative, while highlighting verifiable signals worth watching.
Two-Stage Recommendation System Solves the Ranking Challenge of Tropical Cyclone Port Impacts
A study proposes a two-stage recommendation framework: first retrieve candidate ports by distance to landfall, then rank by AIS impact days. In Hurricanes Nicole and Ian tests, Wide & Deep ranking consistency was 85.5% and 55.5% higher than a distance baseline, helping port authorities allocate emergency resources.
Why You Should Download Your Favorite YouTube Videos
Hacker News debate: Why you should download your favorite YouTube videos? Exploring the fragility of streaming content, copyright disputes, and personal backup practices in the digital age.
Permanent and Transient Representations: Solving the Stability-Plasticity Dilemma in Continual Reinforcement Learning
Continual RL faces a stability-plasticity dilemma. A new study proposes complementary parametric permanent and non-parametric transient value functions, achieving better online learning on image tasks and Craftax-Classic and offering new ideas to mitigate catastrophic forgetting.
Selective Retrieval Routing: A New Paradigm for Multi-Violation Knowledge Graph Repair
This article introduces a selective retrieval routing framework for multi-violation knowledge graph repair, using an LLM repair agent and dynamic router to control external evidence access, achieving more efficient repair on building automation datasets while balancing repair quality and token consumption.