Reading desk

What arrived, what needs you.

Everything across 13 topics — what's new, what needs you, what to read.

13Topics

133Sources

54Papers

22New / 7d

12Attention

60 to read, ranked by priority

P1
The Harness Effect: How Orchestration Design Sets the Token Economics of Enterprise Agentic AINew
Harnesses & Agent SystemsPreprintSep 11
P1
Less Context, Better Agents: Efficient Context Engineering for Long-Horizon Tool-Using LLM AgentsNew
Harnesses & Agent SystemsPreprintSep 11
P1
FlashAttention-4: Algorithm and Kernel Pipelining Co-Design for Asymmetric Hardware ScalingNew
Compilers & KernelsPreprintSep 11
P1
Benchmarking the Benchmarks: A Validity Audit of Tool-Calling EvaluationNew
Evaluation & MeasurementPreprintSep 11
P1
An Interpretable Latency Model for Speculative Decoding in LLM ServingNew
Serving & RuntimePreprintSep 11
P1
ContextLeak: Exfiltrating LLM Agent Context via Malicious ToolsNew
Agent SecurityPreprintSep 11
P1
Breaking the Protocol: Security Analysis of the Model Context Protocol Specification and Prompt Injection Vulnerabilities in Tool-Integrated LLM AgentsNew
Agent SecurityPreprintSep 11
P1
Quantifying construct validity in large language model evaluationsNew
Evaluation & MeasurementPreprintSep 11
P1
Next-Level Inference: Why Your Single-Node vLLM Setup Needs Prefill-Decode DisaggregationNew
Serving & RuntimeTech ReportSep 11
P1
Kimi K3: Open Frontier IntelligenceNew
ModelsTech ReportSep 11
P1
TSMC — 6-K: August 2026 Revenue Report (filed 2026-09-10)New
Hardware — Memory & Compute ArchitectureTech ReportSep 11
P1
SK hynix Announces 2Q26 Financial ResultsNew
Hardware — Memory & Compute ArchitectureTech ReportSep 11
P1
Measuring energy and water efficiency for Microsoft datacentersNew
Datacenters & Digital InfrastructureTech ReportSep 11
P1
ERCOT Market Notice M-A080326-01 — Update Regarding Batch Zero Timelines and ProcessesNew
Energy & PowerTech ReportSep 11
P1
EIA expects record electricity generation in 2026 and 2027 (September 2026 STEO)New
Energy & PowerTech ReportSep 11
P1
CoreWeave Closes $2.6 Billion Loan Facility (DDTL 5.5)New
Capital StructureTech ReportSep 11
P1
CoreWeave Closes $3.1 Billion Loan Facility (DDTL 5.0)New
Capital StructureTech ReportSep 11
P1
ReQAT: Achieving Full-Precision Reasoning Accuracy with 4-bit Floating-Point Quantization-Aware TrainingNew
Compilers & KernelsConferenceSep 11
P1
What Authors Need to Know About the Anthropic Settlement (Bartz v. Anthropic, $1.5bn class settlement)New
DataAnalysisSep 11
P1
Inside Microsoft's two-decade push to cut water intensity while scaling for growthNew
Datacenters & Digital InfrastructureAnalysisSep 11
P2
NVIDIA Announces Spectrum-X Photonics, Co-Packaged Optics Networking Switches to Scale AI Factories to Millions of GPUsNew
Networking & InterconnectNewsSep 11
P2
Broadcom Announces Tomahawk 6 - Davisson, the Industry's First 102.4-Tbps Ethernet Switch with Co-Packaged OpticsNew
Networking & InterconnectNewsSep 11
P2
Rethinking the Evaluation of Harness Evolution for AgentsRead in full
Harnesses & Agent SystemsPaperJul 20
P3
Stop Comparing LLM Agents Without Disclosing the Harness
Harnesses & Agent SystemsPaperJul 23
P3
Harness-Bench: Measuring Harness Effects across Models in Realistic Agent Workflows
Harnesses & Agent SystemsPaperJul 23
P3
Model-First Reasoning LLM Agents: Reducing Hallucinations through Explicit Problem Modeling
Harnesses & Agent SystemsPaperApr 15
P3
Mapping the Exploitation Surface: A 10,000-Trial Taxonomy of What Makes LLM Agents Exploit Vulnerabilities
Harnesses & Agent SystemsPaperApr 14
P3
Your Agent, Their Asset: A Real-World Safety Analysis of OpenClaw
Harnesses & Agent SystemsPaperApr 14
P3
The Amazing Agent Race: Strong Tool Users, Weak Navigators
Harnesses & Agent SystemsPaperApr 14
P3
Agentic AI Security & Autonomous Red-Teaming
Harnesses & Agent SystemsPaperApr 5
P3
Memory for Autonomous LLM Agents: Mechanisms, Evaluation, and Emerging Frontiers
Harnesses & Agent SystemsPaperApr 5
P3
A-MEM: Agentic Memory for LLM Agents
Harnesses & Agent SystemsPaperApr 5
P3
Agentic Reasoning for Large Language Models
Harnesses & Agent SystemsPaperApr 5
P3
From LLM Reasoning to Autonomous AI Agents: A Comprehensive Review
Harnesses & Agent SystemsPaperApr 5
P3
Agentic Tool Use in Large Language Models
Harnesses & Agent SystemsPaperApr 5
P3
How Inference Compute Shapes Frontier LLM Evaluation
Evaluation & MeasurementPaperJul 23
P3
Speculative Decoding Meets Quantization: Compatibility Evaluation and Hierarchical Framework Design
Serving & RuntimePreprintJul 24
P3
Reformulating KV Cache Eviction Problem for Long-Context LLM Inference (LaProx)
Serving & RuntimePreprintJul 24
P3
Measuring Harness-Induced Belief Divergence in Multi-Step LLM Agents
Harnesses & Agent SystemsPreprintJul 23
P3
VeriCache: Turning Lossy KV Cache into Lossless LLM Inference
Serving & RuntimePreprintJul 23
P3
C2KV: Compressed and Composable KV Cache Reuse for Efficient LLM Inference
Serving & RuntimePreprintJul 23
P3
AMMA: A Multi-Chiplet Memory-Centric Architecture for Low-Latency 1M Context Attention Serving
Serving & RuntimePreprintJul 23
P3
StreamDQ: Near-Memory Weight DeQuantization in Custom HBM for Scalable AI Inference Acceleration
Serving & RuntimePreprintJul 23
P3
DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation
Serving & RuntimePreprintJul 23
P3
Spec-AUF: Accept-Until-Fail Training under Train-Inference Misalignment for Masked Block Drafters
Serving & RuntimePreprintJul 23
P3
Quantize the Target, Quantize the Drafter: Efficient Inference with Qwen3.5-4B
Serving & RuntimePreprintJul 23
P3
OSAQ: Outlier Self-Absorption for Accurate Low-bit LLM Quantization
Serving & RuntimePreprintJul 23
P3
From Tensor Buffer to Distributed Memory Hierarchy: A Survey of KV Cache Management for LLM Serving
Serving & RuntimePreprint
P3
TokenStack: A Heterogeneous HBM-PIM Architecture and Runtime for Efficient LLM Inference
Serving & RuntimePreprint
P3
Pi — a minimal, extensible terminal coding harness
Harnesses & Agent SystemsToolAug 6
P3
Can a Language Model Learn Facts Continually in Its Weights?
Harnesses & Agent SystemsTech ReportAug 6
P3
Prime Agent: A self-improving RLM harness
Harnesses & Agent SystemsTech ReportAug 6
P3
verifiers v1: Decomposing Tasksets and Harnesses for Agentic RL & Evaluations
Harnesses & Agent SystemsTech ReportAug 6
P3
AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms
Harnesses & Agent SystemsTech ReportApr 5
P3
Is a Compute Crunch Coming? (inference serving-capacity model, calibrated to SemiAnalysis InferenceX)
Serving & RuntimeTech ReportJul 23
P3
LLM Inference Handbook
Inference & Token-Pricing EconomicsTech ReportAug 6
P3
Intel Q2 2026 Results (Actual Print) — 8-K Ex-99.1
Hardware — Memory & Compute ArchitectureTech ReportJul 24
P3
Prefill/Decode-Aware Evaluation of LLM Inference on Emerging AI Accelerators
Serving & RuntimeConference
P3
Harness Engineering for Self-Improvement
Harnesses & Agent SystemsAnalysisAug 6
P3
Power 2026 — a primer on electricity pricing and AI datacenter development
Datacenters & Digital InfrastructureAnalysisAug 6