Models & Research
How to Stop an LLM Agent From Repeating Failed Steps
Simulations show separating stopping authority from action generation cuts token spend by 36.4% without reducing success rates in LLM agents.
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 and llama.cpp both run quantized LLMs on Apple Silicon unified memory. Here is what each project documents, and how to measure which wins on your Mac.
Models & ResearchChinese AI Models Compared: DeepSeek, Qwen, Kimi, Doubao, and Ernie
Compare DeepSeek, Qwen, Kimi, Doubao, ERNIE and GLM through dated benchmarks, license terms, context windows, API pricing and practical workload fit.
Ethics, Policy & SafetyWhy Self-Improvement Claims Need Benchmarks the Agent Has Never Seen
Self-improvement claims require held-out benchmarks, lineage logs, and calibrated judges to distinguish genuine capability gains from memorized feedback in agent loops.
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.
Open SourceKeep Android Open: F-Droid's Fight Against a Locked-Down Mobile Future
F-Droid, the open-source Android app repository, is leading a global campaign against Google's mandatory developer verification program, a policy set to take effect in September 2026 that critics say will end alternative app distribution and hand Google total control over what software can run on Android devices.
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.
Explore the topics
Guides, comparisons and analysis, organized by topic.
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 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.
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.
The economics, interop standards, and workflow tradeoffs reshaping how code gets written, reviewed, and shipped when AI agents share the editor with the engineer.
More recent analysis
Ethics, Policy & SafetyAI Monoculture: Does Homogenization Change What We Say or Just How?
A 2026 preprint reports AI homogenizes code syntax but not intent, while algorithmic monoculture models show bounded welfare harm, requiring distinct policy levers.
Culture & SocietyWhen AI Coverage Says ChatGPT 'Thinks,' Does Public Perception Shift?
An 815-participant experiment found anthropomorphic AI verbs barely shift perception, while explicit danger framing does. Style guides should prioritize risk accuracy over ban
Industry & BusinessPricing LLM Training Data: Measured Gain vs Flat Token Rates
A new preprint proposes pricing LLM training data by measured gain rather than flat token rates, offering a hybrid contract structure for buyers and sellers.
SecurityScoping AI Agent Access: Can LLM-Inferred Permissions Stay Auditable?
MetaPermit proposes LLM-inferred attributes for auditable agent access control, reporting 31% consistency gains in a self-reported preprint.
Industry & BusinessAI Research Agents vs Kaggle Experts: What the Version Logs Show
TraceML pairs human and agent Kaggle logs to show agents collapse into narrow loops, supporting trajectory audits over leaderboard scores for autonomy decisions.
Agents & FrameworksMCP Agents Cite the Wrong Source: Pooled vs Source-Aware Fact Checks
Pooled faithfulness checkers miss cross-source conflation in MCP agents. ProvenanceGuard adds per-claim source verification, though author-reported results remain unreplicated
Open SourceCloudflare Forge vs OpenAPI Generator: Choosing an SDK Pipeline for AI Agents
Cloudflare Forge is a new open-source SDK pipeline for AI agents. Evidence is limited to vendor claims, so pilot one artifact before migrating from incumbent generators.
Agents & FrameworksCloudflare's cf CLI: Which API Permissions Should a Coding Agent Get
Cloudflare's cf CLI expands agent access to 3,000 operations. Grant read-only scopes by default and gate mutations behind human checkpoints to limit blast radius.
- Developer Tools
Cutting LLM Prompt Tokens: When Fidelity Survives the Trim
A preprint suggests trimming LLM prompts saves tokens, but fidelity checks are weak and cached input pricing reduces the financial benefit of removing redundant context.
Open SourceESLint vs tsgolint: When Faster Linting Pays Off for Coding Agents
Vendor-reported data suggests tsgolint is faster than ESLint, but teams should profile agent loops and pilot one repo to verify gains before migrating CI.
Ethics, Policy & SafetyShould Therapists Use LLMs on Session Transcripts? What the Evidence Supports
Evidence supports restricting LLM use on therapy transcripts to clinician-reviewed drafting, as the sole preprint lacks outcome data and shows inconsistent rater agreement.
Open SourceMermaid vs draw.io: Choosing Auto-Layout or Manual Diagram Placement
For docs-as-code repos, default to Mermaid for auto-layout diagrams, reserve draw.io for exact placement, and pilot Reladraw's relative syntax as an experimental middle ground
From the archive
Featured analysis and deeper reads.