Groundy · Most read

MLX vs llama.cpp on Apple Silicon: Which Runtime to Use for Local LLM Inference

Berry Mingus — Tue, 24 Mar 2026 16:48:05 GMT

MLX delivers 20-87% faster generation on Apple Silicon for models under 14B parameters. llama.cpp wins for cross-platform use and long contexts.

Berry Mingus · Infrastructure & Runtime · 9 min read

Read the full article on Groundy →

EU's 2027 Replaceable Battery Mandate: What It Means for Phone Buyers and Repairers Right Now

Berry Mingus — Tue, 21 Apr 2026 16:00:00 GMT

The EU's 2027 battery mandate is confirmed. Here's what 'user-replaceable' legally means, which phones comply now, and how to buy smart before the rules change.

Berry Mingus · Culture & Society · 6 min read

Read the full article on Groundy →

Open-Weight LLM Leaderboards 2026: Where DeepSeek, Qwen, and GLM Rank

Berry Mingus — Fri, 26 Jun 2026 22:09:15 GMT

DataLearner's June 2026 snapshot ranks GLM-5.2 seventh by HLE at 54.70 and places no Chinese flagship in the overall top three, undercutting launch-day claims.

Berry Mingus · Models & Research · 8 min read

Read the full article on Groundy →

Chinese AI Models Compared: DeepSeek, Qwen, Kimi, Doubao, and Ernie

Berry Mingus — Tue, 24 Mar 2026 14:57:11 GMT

DeepSeek isn't China's only frontier AI. Compare DeepSeek, Qwen, Kimi, Doubao, and Ernie on benchmarks, licensing, API access, and use-case fit.

Berry Mingus · Models & Research · 10 min read

Read the full article on Groundy →

4-Bit vs 8-Bit Quants: Why Accuracy Benchmarks Miss Distribution Drift

Berry Mingus — Sat, 12 Sep 2026 04:11:48 GMT

A preprint argues zero-shot accuracy misses distribution drift in quantized LLMs, recommending divergence metrics like JSD and TV against BF16 bases for safer deployment.

Berry Mingus · Developer Tools · 9 min read

Read the full article on Groundy →

FP8 vs MXFP4 vs BF16: Why Your Quantized LLM Disagrees Across GPUs

Berry Mingus — Mon, 07 Sep 2026 07:15:12 GMT

FP8 and MXFP4 are umbrella specs, not single formats. A new preprint offers bit-exact conformance vectors to test quantized LLM portability across GPUs, exposing hidden format

Berry Mingus · Models & Research · 13 min read

Read the full article on Groundy →

GitHub Copilot vs Cursor vs Claude Code: The 2026 AI Coding Showdown

Berry Mingus — Sat, 14 Mar 2026 16:29:11 GMT

GitHub Copilot owns enterprise, Cursor owns developer wallets at $2B ARR, and Claude Code leads the benchmarks. Which fits your workflow depends on what you build.

Berry Mingus · Developer Tools · 11 min read

Read the full article on Groundy →

Running MoE LLMs on a Single GPU: What Expert Offloading Actually Costs

Berry Mingus — Fri, 11 Sep 2026 20:14:36 GMT

FluxMoE streams MoE weights from host DRAM to save VRAM, trading capacity for bandwidth. Author-reported gains on multi-GPU setups lack independent replication.

Berry Mingus · Infrastructure & Runtime · 9 min read

Read the full article on Groundy →

Private Vector Search vs TEEs: Can RAG Retrieval Be Outsourced Safely?

Berry Mingus — Sun, 06 Sep 2026 08:06:35 GMT

Spruce claims 0.21-2.97s private retrieval via MPC, challenging TEE defaults. Compare self-hosted Qdrant, TEEs, and cryptographic outsourcing for privacy-sensitive RAG.

Berry Mingus · Infrastructure & Runtime · 13 min read

Read the full article on Groundy →

Cursor's Meteoric Rise: Inside the AI Editor Hitting $300M ARR

Berry Mingus — Fri, 27 Feb 2026 14:15:59 GMT

Cursor hit $300M ARR in April 2025 by forking VS Code and baking AI into the editor's core. By June 2026 it was at $4B annualized and agreed to a $60B SpaceX acquisition. Here's how it happened and what it signals.

Berry Mingus · Industry & Business · 8 min read

Read the full article on Groundy →

AI Code Generation Benchmarks 2026: Which Model Actually Writes Better Code?

Berry Mingus — Sun, 15 Feb 2026 11:01:06 GMT

Frontier and open-weight coding models now post similar benchmark scores, but real-world software engineering exposes gaps between leaderboard results and practical utility.

Berry Mingus · Models & Research · 9 min read

Read the full article on Groundy →

Constitutional AI Self-Amendment Hits the Metacognition Wall

Berry Mingus — Sun, 13 Sep 2026 08:57:23 GMT

An arXiv preprint argues LLM metacognition is coarse and context-dependent, suggesting self-amendment requires external calibration and human approval gates.

Berry Mingus · Ethics, Policy & Safety · 8 min read

Read the full article on Groundy →

Do More Tools Make LLM Agents Worse? When to Gate External Evidence

Berry Mingus — Fri, 11 Sep 2026 20:58:31 GMT

Preprints show forced evidence gathering raised clinical agent error from 28.3% to 34.3%, while curated tool menus lifted ToolBench success to 0.898, suggesting gating is key.

Berry Mingus · Agents & Frameworks · 9 min read

Read the full article on Groundy →

Terminal vs Browser for AI Agents: What a Hybrid GUI+CLI Benchmark Shows

Berry Mingus — Mon, 07 Sep 2026 19:36:21 GMT

CUA-Universe preprint shows hybrid GUI+CLI agents cut steps 37% and tokens 60% on 16 apps. Audit your eval stack: screenshot-only benchmarks mismeasure tasks with terminal.

Berry Mingus · Developer Tools · 14 min read

Read the full article on Groundy →

How Cloudflare Detects MCP Traffic and What Enterprises Should Filter

Berry Mingus — Mon, 07 Sep 2026 08:56:55 GMT

Cloudflare claims MCP traffic classification, but stdio servers remain invisible to edge proxies. Build a filter policy that gates remote endpoints and treats local tool on as

Berry Mingus · Agents & Frameworks · 11 min read

Read the full article on Groundy →

Noisy Neighbors at the Fabric: Why Shared GPU Clusters Throttle Your Jobs

Berry Mingus — Mon, 07 Sep 2026 07:59:55 GMT

A new NCCL shim recovers 13-38% bandwidth on shared GPU clusters by tuning collective patterns. Test for cross-tenant interference before buying more fabric.

Berry Mingus · Infrastructure & Runtime · 12 min read

Read the full article on Groundy →

How Spotify Cut Claude Code Token Usage 90%: What It Takes to Replicate

Berry Mingus — Sat, 05 Sep 2026 20:04:28 GMT

Spotify's claimed 90% token cut via Portal is unverified. Benchmark token proxies against Claude Code's native compaction and subscription ceilings before building.

Berry Mingus · Agents & Frameworks · 12 min read

Read the full article on Groundy →

Claude Code in GitHub Actions: A Complete Guide to Automated PR Fixes

Berry Mingus — Tue, 24 Mar 2026 14:36:19 GMT

How to wire Claude Code into GitHub Actions for automated PR fixes, CI failure remediation, and code review, with cost controls, model options, and security guardrails.

Berry Mingus · Developer Tools · 10 min read

Read the full article on Groundy →

Google's TimesFM: A Foundation Model for Time Series

Berry Mingus — Fri, 27 Feb 2026 16:43:07 GMT

TimesFM is Google's decoder-only transformer for zero-shot time-series forecasting, pretrained on ~100 billion real-world points across domains without retraining.

Berry Mingus · Models & Research · 9 min read

Read the full article on Groundy →

WiFi DensePose: Full-Body Tracking Through Walls Using Your Router

Berry Mingus — Wed, 18 Feb 2026 12:20:50 GMT

WiFi routers can perform full-body pose estimation through walls using Channel State Information, turning everyday network infrastructure into a covert tracking system.

Berry Mingus · Models & Research · 6 min read

Read the full article on Groundy →