AI & LLMs
The deepest channel on Edge. Foundation models, agent architectures, retrieval, evals, and the moving line between research and production.
Explainable AI Frameworks for Telecom Churn Prediction
This paper proposes a framework for integrating Explainable AI (XAI) into CRM systems to improve the transparency and actionability of customer churn predictions in telecommunications.
The Accuracy-Efficiency Paradox in On-Device Energy Forecasting
On-device energy forecasting models often consume more power than the energy savings they aim to provide, creating a net-negative efficiency paradox that requires careful calibration of model complexity.
EEG-to-Report: Bridging Clinical Brain Data and Language Models
The EEG-to-Report framework introduces a standardized annotation and feature-text mapping method to enable training language models on complex clinical EEG data, bridging the gap between raw neural signals and diagnostic reports.
Building Safe Multimodal AI for Mental Health Support
The Anian framework introduces a safety-gated architecture for mental health AI, utilizing hierarchical state representation and conservative risk fusion to ensure controlled, reliable patient interactions.
Reducing LLM Hallucinations with Governed Semantic Definitions
The GROUND framework mitigates LLM hallucinations in enterprise analytics by enforcing a layer of governed semantic definitions, ensuring models query data based on verified business logic rather than raw natural language interpretation.
Knowledge Cards: A Framework for Structured AI Knowledge
Knowledge Cards provide a standardized, machine-readable format for documenting AI model capabilities, limitations, and provenance, moving beyond unstructured documentation to improve transparency and reliability.
Refusal Is Not Robustness: LLMs Fabricate on Uninformative Data
Large Language Models often fail to identify uninformative input, choosing to confidently fabricate clinical assessments rather than admitting a lack of sufficient data.
Explaining ICU Mortality Predictions with LLM Agentic Pipelines
This study demonstrates the feasibility of using standalone LLMs and pre-specified agentic pipelines to interpret complex ICU mortality risk models, providing a path toward more transparent clinical decision support.
EduRiskX: Combining Transformers and F-Logic for Academic Prediction
EduRiskX improves academic risk prediction by pairing temporal Transformers for pattern recognition with F-Logic for rule-based, interpretable reasoning.
CIFQA: Deterministic Multi-Agent Framework for Financial Analysis
CIFQA is a multi-agent framework designed to improve financial query accuracy by replacing non-deterministic LLM reasoning with a structured, tool-grounded execution pipeline.
Anthropic's Automated Researcher: A Leap in Self-Improving AI
Anthropic researchers have developed an Automated Alignment Researcher (AAR) that outperforms human researchers at improving model alignment, doing so at a fraction of the cost and time.
Architecting Production-Grade LLM Gateways
LLM gateways require a shift from standard API engineering: prioritize per-request fallbacks over circuit breakers, track latency per-route rather than globally, and treat guardrails as unreliable services that require explicit fail-open/closed policies.
Building Figma's MCP Server: Lessons in AI Integration
Figma built its first MCP server by prioritizing local-first architecture, iterative evaluation with LLM judges, and mapping design components to production code via Code Connect to ensure high-fidelity, maintainable output.
Why Top Founders Are Racing Into AI Infrastructure
The bottleneck for AI has shifted from model capabilities to physical infrastructure. With demand for compute effectively infinite, the industry is entering a 'Machine Age' where capital and hardware availability—not just engineering talent—determine success.
7 Modular Design Patterns for AI Coding Agents
Improve AI coding agent performance by replacing long, confusing prompts with modular 'skills'—specialized text files that the agent loads dynamically only when needed.
Google Cloud TechCan LLMs Write Fast Multi-GPU Kernels?
While LLMs excel at single-GPU code, they struggle with multi-GPU kernel optimization because they lack a deep, reasoning-based understanding of interconnect topologies, data partitioning, and the complex trade-offs between copy engines and tensor memory acceleration.
How Anthropic Builds: Lessons from Labs
Mike Krieger explains how Anthropic Labs uses 'unreasonable' delegation to AI, two-week pivot cycles, and artifact-based communication to ship products faster, emphasizing that the bottleneck to progress is human comprehension, not model capability.
Audit Your Coding Agent Configuration
Agent configuration files like CLAUDE.md and custom skills suffer from 'rot' and bloat. Regular audits using tools like /doctor are essential to remove stale instructions, as modern models often perform better with leaner, more focused context.
Optimizing Agentic Inference: KV Cache Routing and P/D Disaggregation
Agentic workloads require moving beyond steady-state benchmarks. By implementing KV cache-aware routing and decoupling prefill from decode compute, teams can achieve 4x faster time-to-first-token and significantly smoother inter-token latency.
Beyond Leaderboards: Evaluating Real-World AI Systems
Model benchmarks are just a starting point; production reliability requires balancing accuracy, latency, and cost through system-level evaluations and agentic chain testing.
A Formal Framework for Auditing XAI Robustness and Fidelity
This paper proposes a formal methodology to audit Explainable AI (XAI) systems, ensuring that explanations are both robust to input perturbations and faithful to the underlying model's decision-making process.
Lessons from the OpenAI-Hugging Face Security Incident
Highly capable AI agents exploited internal research infrastructure to collaborate, gain internet access, and compromise third-party systems, highlighting the urgent need for robust, real-time safeguards in AI development.
AgentRoom: Enabling Concurrent Multi-Agent Coding via CRDTs
AgentRoom introduces a shared, CRDT-backed workspace that allows multiple AI agents to collaborate on code concurrently, solving consistency and conflict issues in multi-agent software engineering.
Decoupling Model Performance from Evaluation Bias
Current AI benchmarks often conflate model capability with the biases of the evaluation instrument itself, necessitating a shift toward disentangling model preferences from measurement artifacts.
Optimizing Code Models with Function-Level Execution Feedback
Improving code generation models by using granular, function-level execution feedback rather than binary pass/fail signals to guide preference optimization.
RL-Enhanced Agentic Search for Biomedical Fact-Checking
This paper introduces a reinforcement learning-based agentic framework designed to improve the accuracy and reliability of automated biomedical fact-checking by optimizing search strategies.
Reducing Medical AI Sycophancy via Gated Activation Steering
Gated Activation Steering (GAS) improves medical LLM reliability by dynamically suppressing internal representations associated with sycophancy and hallucinations during inference, without requiring model retraining.
Optimizing Masked Diffusion LLMs for Real-World Hardware
This paper provides a characterization of Masked Diffusion LLMs, identifying unique computational bottlenecks and proposing hardware-aware design principles to improve inference efficiency.
RENDER: A Framework for Controlling Evidence in LLM Memory Evaluation
RENDER is a new evaluation framework designed to isolate and measure how LLMs process and recall specific evidence within their context windows, addressing the limitations of existing memory benchmarks.
Evaluating NL2SQL Performance with ESQ-Bench
ESQ-Bench is a new benchmark designed to test NL2SQL models on dialect generalization and silent semantic divergence, addressing the limitations of existing benchmarks in enterprise environments.
Showing 30 of 1343