groundy

Groundy — independent coverage of developer tools, infrastructure, and platforms

models

DeepSeek 32B on RTX 3090: Tokens per Second by Quant and Context

Bandwidth math predicts 32-39 tok/s for Q4_K_M DeepSeek 32B on an RTX 3090 at 8k context. 32k context overflows 24GB VRAM. No measured benchmarks exist, so every figure is.

6min·21:50·aug 21
models

PTXBench: LLMs Can Port GPU Kernels, But Not Beat Tuned Libraries

PTXBench shows LLMs can generate architecture-specific PTX for H100 and B200, but no model matches frontier libraries. Use it to cut hot-loop porting costs, not to replace.

6min·14:08·aug 21
agents

Vibe Coding vs Control: How Developers Actually Used AI Coding Agents

New data shows pro devs use AI coding agents as collaborators, not delegates. Weight review surfaces and checkpoints over autonomous run length when buying agent tooling.

policy

Android Privacy Policies vs. Runtime Logs: A 0.4% Alignment Study

A study of 1,000 Android apps found only 0.4% align privacy policies with runtime logs. 67.6% leak sensitive data. Shift compliance from document review to CI log-inventory.

industry

LLM Data Center Control: Why Advisory Beats Closed-Loop

SemPlan data shows LLM planners fail on 74% of operational queries. Keep actuation in classical MPC, use LLMs only for advisory, and demand third-party verification before.

infra

V8 Isolates vs MicroVMs vs Wasm: Where Spectre Still Draws the Line

Cloudflare's 2026 Spectre revisit is unverified. This guide maps V8 isolates, Firecracker microVMs, and Wasm sandboxes to define a data-classification rule for edge secrets.

models

Can LLMs Reuse Another Model's KV Cache? What Cross-Model Transfer Shows

A new preprint proposes a lightweight reader to reuse KV caches across models, but evidence suggests caches remain model-bound. Learn when recompute beats adaptation.

infra

Fine-Tuning DeepSeek Without NVIDIA: What the Ascend SuperPOD Run Shows

SLAI T-Rex reports 34.22% MFU for full-parameter DeepSeek-V4 post-training on Ascend. The single-cluster result narrows the CUDA moat for fine-tuning, but cost data remains.

featurededitor's picks
most popularlast 6 days · by traffic
tophighest rated
feedall 681 articles · reverse chronological
  1. aug 21modelsDeepSeek 32B on RTX 3090: Tokens per Second by Quant and Context
  2. aug 21modelsPTXBench: LLMs Can Port GPU Kernels, But Not Beat Tuned Libraries
  3. aug 21agentsVibe Coding vs Control: How Developers Actually Used AI Coding Agents
  4. aug 21policyAndroid Privacy Policies vs. Runtime Logs: A 0.4% Alignment Study
  5. aug 21industryLLM Data Center Control: Why Advisory Beats Closed-Loop
  6. aug 21infraV8 Isolates vs MicroVMs vs Wasm: Where Spectre Still Draws the Line
  7. aug 20modelsCan LLMs Reuse Another Model's KV Cache? What Cross-Model Transfer Shows
  8. aug 20infraFine-Tuning DeepSeek Without NVIDIA: What the Ascend SuperPOD Run Shows
  9. aug 20agentsMulti-Agent or Single-Agent LLM: What Skill Distillation Actually Costs
  10. aug 20devtoolsTraining a Personal Coding Assistant: GPU Cost vs a Copilot Seat
  11. aug 20agentsMulti-Agent LLM Systems Drift Into Misaligned Communication Over Long Horizons
  12. aug 19infraCloudflare WebMCP: The Security Baseline for Agent-Ready Sites
  13. aug 19policyWhy Machine Unlearning Can't Certify GDPR Erasure
  14. aug 19agentsSizing Agent Memory: A Capacity Planning Rubric for Long-Horizon LLMs
  15. aug 19infraCloudflare AI Search vs Self-Hosted RAG: Where the Build-vs-Buy Line Lands
  16. aug 18devtoolsVCoT-Bench: Why AI Rust Verification Fails Merge Gates
  17. aug 18infraCloudflare H1 2026 DDoS Report: DNS Floods and Sizing Past 1 Tbps
  18. aug 18industryLLM Conflict-of-Interest Benchmark: Sponsor Bias as a Measurable Failure Mode
  19. aug 18policyRA-Bench: Why Deepfake Detectors Fail on Re-Shared Crisis Video
  20. aug 18agentsWhen Spec-First Agents Dismantle Invariants: A Governance Case Study
  21. aug 18devtoolsRust GPU Offload: arXiv 2608.13759 Analysis for Rust Teams
  22. aug 18agentsCloudflare Kitesurf: V8 Isolates vs Containers for Agent Browsing
  23. aug 18policyDario Amodei on AI Regulation: Frontier Labs as Their Own Lobbyists
  24. aug 18agentsClaude Code Skills vs Model Weights: Where Should Agent Skills Live?
  25. aug 17devtoolsCursor Origin vs GitHub: The Real Cost of Switching Repo Hosts
  26. aug 17devtoolsTreat AI Autofix as Untrusted Input: Merge Gates and CI Scoping
  27. aug 01modelsDeCRIM: Decompose Constraints to Stop Silent Drops in Agent Outputs
  28. aug 01devtoolsKimi K3 Local Inference: Why 2.8T Parameters Break the Consumer RAM Floor
  29. jul 31modelsOperator-Level Triage for Silent Mixed-Precision Instability
  30. jul 31modelsBeyondUncertainty: Weak Confidence Signal for RAG Routing, Not Calibration
  31. jul 31policyPublic Sector AI Procurement Must Shift from Model Certification to Task Authorization
  32. jul 30infradaVinci-kernel shifts the RL kernel bottleneck from reward shaping to skill libraries
  33. jul 30agentsCloudflare Precursor: Behavioral Detection for AI Agents
  34. jul 30devtoolsProvenance as a CI Gate: Attributing Agent-Authorship in Code
  35. jul 30devtoolsOHTTP CLI: Stateless Privacy for Agents vs VPN and Tor
  36. jul 30modelsKimi K3 on M1 Max: Bandwidth, Not Capacity, Limits Local MoE Inference
  37. jul 30policyWhy Written AI Policies Fail to Control Agent Behavior
  38. jul 29agentsWhy Multi-Agent LLM Delegation Concentrates Risk
  39. jul 29modelsKimi Linear Cuts KV Cache 75% but Recall Remains the Binding Constraint
  40. jul 29policyWhy Vendor Model Cards Fail Clinical Ethics Procurement
  41. jul 29agentsHarness vs Scaffold: Why Claude Code and LangGraph Are Not Interchangeable
  42. jul 28devtoolsFine-Tuning vs RAG for Internal APIs: StarCoder2 Constraints
  43. jul 28agentsMCP Tool Discovery Moves From Hardcoded Config to Runtime Agent Search
  44. jul 28infraCalibrated LLM Monitoring: Conformal Prediction with Drift Detection
  45. jul 28devtoolsMellum2 Unverified: Why MoE Active Parameters Matter More Than Total Size
  46. jul 28agentsCode-as-Action Agents Beat GAIA But Require Runtime Sandboxing
  47. jul 28policyTRIDENT Benchmark: LLM Safety Gaps in Finance, Medicine, and Law
  48. jul 28modelsWhy Chat Leaderboards Do Not Predict Image Quality
  49. jul 27devtoolsPyPI Wheel Reproducibility: 15% Byte-Identical, 79% Source-Equivalent
  50. jul 27modelsKimi K3 Procurement: Governance Review Over Phantom Government Assessments
load older →