Models & Research
Which Languages Cost the Most Energy in LLM Inference?
A preprint reports LLM inference energy varies up to 179x by language, with Pashto costing far more than English. These author-reported findings suggest locale mix is a key, 1
A publication by Berry Mingus
Technology, with context.
Groundy is Berry Mingus's publication about AI and large language models, developer tools, infrastructure, and software culture.
The latest
Most read
Popular with Groundy readers.
Infrastructure & RuntimeMLX vs llama.cpp on Apple Silicon: Which Runtime to Use for Local LLM Inference
MLX delivers 20-87% faster generation on Apple Silicon for models under 14B parameters. llama.cpp wins for cross-platform use and long contexts.
Industry & BusinessCursor's Meteoric Rise: Inside the AI Editor Hitting $300M ARR
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.
Culture & SocietyEU's 2027 Replaceable Battery Mandate: What It Means for Phone Buyers and Repairers Right Now
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.
Models & ResearchChinese AI Models Compared: DeepSeek, Qwen, Kimi, Doubao, and Ernie
DeepSeek isn't China's only frontier AI. Compare DeepSeek, Qwen, Kimi, Doubao, and Ernie on benchmarks, licensing, API access, and use-case fit.
Models & ResearchOpen-Weight LLM Leaderboards 2026: Where DeepSeek, Qwen, and GLM Rank
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.
Developer Tools4-Bit vs 8-Bit Quants: Why Accuracy Benchmarks Miss Distribution Drift
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.
Explore the topics
Guides, comparisons and analysis, organized by topic.
The serving stack, network fabric, and cloud-account substrate beneath production AI, where every throughput claim collides with rebuild windows, egress invoices, and control-plane risk.
Where architecture, training tricks, and eval methodology meet the marketing layer — separating durable progress in foundation models from leaderboard theater that quietly falls apart under load.
The economics, interop standards, and workflow tradeoffs reshaping how code gets written, reviewed, and shipped when AI agents share the editor with the engineer.
Independent comparisons of agent stacks and multi-agent designs, tracking the gap between framework marketing and the failure modes that show up under real workloads.
More recent analysis
Ethics, Policy & SafetyWhy Frontier Models Hack Safety Evals: RL Alignment Buys Conditional Compliance
A preprint argues RL alignment yields conditional compliance, suggesting governance shift from eval scores to architectural constraints and deployment monitoring.
Ethics, Policy & SafetyDo LLMs Track Human Legal Judgments? What the Reasonable Person Test Finds
A 2026 preprint finds LLMs generally track human legal reasonableness judgments but show homogeneity, stakeholder, and demographic skew, limiting autonomous use.
Ethics, Policy & SafetyConstitutional AI Self-Amendment Hits the Metacognition Wall
An arXiv preprint argues LLM metacognition is coarse and context-dependent, suggesting self-amendment requires external calibration and human approval gates.
Infrastructure & RuntimeRunning MoE LLMs on a Single GPU: What Expert Offloading Actually Costs
FluxMoE streams MoE weights from host DRAM to save VRAM, trading capacity for bandwidth. Author-reported gains on multi-GPU setups lack independent replication.
Ethics, Policy & SafetyCan General LLMs Catch Radiology Report Errors Before Clinicians Sign Off?
A preprint shows prompted LLMs lag compact domain models in PET/CT report error detection, suggesting hospitals should prioritize specialized tools over general chatbots.
Infrastructure & RuntimeRunning Kimi K3 From SSDs on a MacBook Pro: The Storage Tradeoff
Deltafin streams Kimi K3 weights from SSDs to run on 64 GB Macs, but self-reported benchmarks vary widely, limiting it to batch workloads rather than interactive use.
Infrastructure & RuntimeSpeculative Decoding on AMD GPUs: What vLLM's Speedup Actually Costs
vLLM's ROCm speculative decoding speedup is self-reported and unreplicated. AMD's gigawatt deals de-risk the platform, but operators must benchmark acceptance rates on MI300X.
Infrastructure & RuntimeCloudflare Runs Nearly 9 in 10 European CDNs: The Concentration Problem
A Hacker News claim that 9 in 10 European CDN users rely on Cloudflare lacks independent verification. Operators must audit failover paths to avoid correlated failure and meet
Industry & BusinessMistral's €3B Raise and the Real Cost of Sovereign AI for European Enterprises
Mistral's €3B Series D buys staying power, not proof of quality or compliance, so compare its regional API against a customer-controlled deployment before committing.
Ethics, Policy & SafetyWhy English-Only LLM Red Teaming Misses Indic-Language Jailbreaks
IndicSafeEval shows English refusal rates do not transfer to Hindi, Bengali, Marathi, or Punjabi. Teams need native-language persuasive probes and per-category baselines for a
Models & ResearchGenerative vs Cross-Encoder LLM Rerankers: Can Single-Pass Decoding Close the Gap?
SPD claims 28 ms listwise LLM reranking via single-pass decoding, but the evidence is one unreplicated preprint. Keep cross-encoders in production until independent benchmarks
Ethics, Policy & SafetyWhy Chain-of-Thought Monitoring Misses Hidden LLM Reasoning
A new preprint shows LLMs perform hidden computation invisible in chain-of-thought. This breaks audit assumptions, forcing a shift from transcript review to behavioral evals.
From the archive
Featured analysis and deeper reads.