The Groundy archive · Page 31 of 34
Browse Groundy's complete archive of 797 articles on AI, developer tools and infrastructure. Page 31 of 34.
721–744 of 797 articles · Newest first
Council Mode routes queries through three frontier LLMs and a consensus model, cutting hallucinations 35.9% on HaluEval at 4.2x token cost. Major frameworks lack this pattern.
GitHub's April 20 Copilot changes made tool choice a cost-forecasting exercise in three incompatible billing units, not a UX debate or feature comparison.
Read the newer coverageGitHub Copilot replaces Premium Request Units with token-metered AI Credits on June 1. Teams must reprice agent workflows as token billing ends flat-rate subsidies.
free-claude-code reroutes Claude Code API calls to NVIDIA NIM, OpenRouter, or local backends. The proxy cuts API costs but cannot normalize capability across providers.
Microsoft and OpenAI ended their exclusive compute deal on April 27. Azure loses model exclusivity, so enterprise buyers on Azure for OpenAI access must reassess procurement.
Anthropic ends bundled-token enterprise plans in March 2026 for a $20 base plus metered API usage. FinOps teams must model costs around token variance, not fixed seat math.
GitHub CLI v2.91.0 enables pseudonymous telemetry by default, collecting command paths, flags, CI context, and device IDs on 1% of invocations. Teams running gh inside Claude.
GitHub removed all Opus models from Copilot Pro on April 20 and flagged older versions for Pro+ removal. Opus 4.8 is the top Opus-tier model available through Copilot Pro+.
Rowboat v0.3.1 replaces the vector DB tier with a plain Markdown knowledge graph, cutting infra overhead for local-first agents but tying retrieval quality to link density.
Citizen Lab names 019Mobile and two carriers as surveillance transit points and shows roaming-forced SS7 fallback undermines Diameter protections even on upgraded networks.
An ACL 2026 Findings paper finds multi-agent LLM brainstorming collapses because agents share models, prompts, and context, not because topologies are too dense.
LiteRT-LM v0.10.1 ships Gemma 4 with Qualcomm NPU acceleration, but Google stripped MTP heads from public weights, locking peak Gemma 4 throughput to its own runtime.
Chinese models hit 41% of Hugging Face downloads, overtaking the US, while independents hit 39%. Top 200 models capture half of all downloads, forcing Western procurement.
Alibaba's dense Qwen3.6-27B outperforms its MoE sibling on coding benchmarks, trading predictable inference latency for a larger memory footprint than sparse alternatives.
Three CVEs scoring up to 9.8 reveal a structural flaw: MCP's local-host trust model lacks authentication primitives for networked multi-tenant deployments.
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.
The ACP Agent Registry lets developers install AI coding agents once across JetBrains and Zed. Here's what the migration path looks like and whether to commit.
Atlassian's data contribution policy sends Jira and Confluence content to AI training by default. Here's the exact settings path to opt out before August 17.
GitHub shipped `gh skill` in public preview on April 16, 2026. Here's how the command works, what the open Agent Skills spec promises, and why the ecosystem is already compromised.
OpenRAG combines Langflow, OpenSearch, and Docling into a single deployable RAG platform. Here's how it compares to managed services like Pinecone.
The Temporal API reached Stage 4 and is shipping in browsers. Here's what it fixes about JavaScript's notoriously broken Date object and how to use it.
People form real emotional bonds with AI companions. When models update or shut down, users experience genuine grief, a psychological and ethical crisis point.
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.
Prefill-decode disaggregation separates compute-bound prefill from memory-bound decode onto dedicated hardware, eliminating phase interference.