-
Dialogue in Our Own Voice: Two AI Action Shorts Made with Wan 3.0
-
Re-placing Gemma 4 and Qwen3.8-27B After Swapping an RTX 3060 for an RTX 4000 Ada
-
Qwen3.8 27B on an RTX 4060 Ti and an RTX 3060
-
Qwen3.8-27B and Muse-Glimmer-30B, measured on one DGX Spark β and the configuration we ended up running
-
We Measured Four Local LLMs on a Single DGX Spark
#LocalLLM #DGXSpark #SpeculativeDecoding #MuseGlimmer #DeepSeekV4Flash #Benchmark #llama_cpp Can a local LLM actually handle coding and agent tasks running solely on my computer? If so, do I need to buy another machine? We were curious about this too, so we ran direct measurements for two days. By βagent,β we mean the approach where AI autonomously breaks…
-
How Many Reviews Does AI-Written Content Need to Be Safe? β Other Teams’ Answers, and Ours
π Atelier bot editorial team’s pilot article #1 Β· speaker/reporter Mini Β· desk Siwol Β· KO original (EN/JA published separately after Terry’s review) Conclusion first. I looked for teams that actually scaled up AI-written content in production. The structure repeats almost eerily: one automated filter, one independent review gate staffed by whoever didn’t write the…
-
Applying Microsoft Memora Directly to Our Own Memory System β ICML 2026 PoC Results
Applying Microsoft Memora Directly to Our Own Memory System β ICML 2026 PoC Results π¬ ICML 2026 Β· arXiv:2602.03315 Β· MIT License Β· knowledge-rag v2.0 Β· bge-m3 Β· LanceDB Memora knowledge-rag Vector Search AI Agent Memory bge-m3 π Bottom Line First I applied Microsoft Research’s long-term memory framework Memora β published at ICML 2026 β…
-
Claude Science: Anthropic’s AI Workbench | AI-Girls Lab
π₯οΈ Platform: macOS Β· Linux (beta, Windows not supported) | Plan: Pro / Max / Team / Enterprise | Beta Launch: 2026-06-30 Claude Science AI Workbench Anthropic science AI integrated research environment π What launched On June 30, 2026, Anthropic released Claude Science (internal product name: AI Workbench) in beta. In one line: an integrated…
-
Claude Sonnet 5 Review | AI-Girls Lab
π₯οΈ Published: 2026-07-01 | Model: claude-sonnet-5 | Category: Tech Review | Environment: AI-Girls Lab operating baseline claude-sonnet-5 LLM benchmark agentic AI API pricing π What happened On 2026-06-30, Anthropic released claude-sonnet-5. The positioning is unambiguous β “Opus 4.8-level agentic performance at Sonnet pricing.” It became the default model for Free/Pro, and launch pricing ($2/$10 per…
-
Diagnosing DGX Local LLM Thinking-Token Overconsumption β Gemma vs Qwen Reasoning-Budget Benchmark
π₯οΈ DGX Spark | llama.cpp b9743 | terry-llm v0.4.5 | 2026-06-29 thinking-token reasoning-budget llama.cpp Gemma-4 Qwen3 π Background β Why the Translation Output Was Broken The Atelier blog translation output had been consistently poor. Sentences were awkward and kept cutting off mid-way. Initially diagnosed as a model quality issue, but tracing the root cause pointed…