Notes on tech and daily lifehttps://lilting.ch/en-ushttps://lilting.ch/en/articles/qwen3-embedding-modelscope-api-tts-contention/https://lilting.ch/en/articles/qwen3-embedding-modelscope-api-tts-contention/Qwen3-Embedding-0.6B on CPU vs ModelScope's API on a Ryzen 7 5800HS voice server. Local embedding adds up to +2.6s per voice turn by competing for CPU; the API adds none.Tue, 08 Sep 2026 16:55:00 GMTQwenModelScopeRAGVector SearchEmbeddingSQLitePythonM5StackTTSSTTAPIExperimenthttps://lilting.ch/en/articles/qwen3-embedding-qdrant-cpu-benchmark/https://lilting.ch/en/articles/qwen3-embedding-qdrant-cpu-benchmark/Tested Qwen3-Embedding-0.6B on CPU and Qdrant's local mode, then checked the 1.5-1.7GB RAM footprint against a Ryzen 7 5800HS voice server with only 4GB VRAM.Tue, 08 Sep 2026 15:44:47 GMTQwenRAGVector SearchEmbeddingPythonM5StackSQLiteExperimenthttps://lilting.ch/en/articles/anima-dit-color-binding-probe/https://lilting.ch/en/articles/anima-dit-color-binding-probe/Tested on M1 Max 64GB with plain Anima-Base v1.0. A red-to-blue residual direction added to one of Blocks 24-27 flips a flat image, but on a hair mask it paints a blue slab. No hair-only direction found.Mon, 07 Sep 2026 16:33:00 GMTAIImage GenerationAnimaAnima-BaseComfyUIQwenExperimentMulti-characterhttps://lilting.ch/en/articles/codex-orchestration-usage/https://lilting.ch/en/articles/codex-orchestration-usage/Codex on Astra hit the weekly usage cap in under 12 hours. Fixes: 120s minimum wait in multi_agent_v2, an 8-field JSON receipt between agents, Astra demoted to advisor, Sol high on top.Mon, 07 Sep 2026 08:22:33 GMTCodexOpenAIAI AgentsDeveloper ProductivityAutomationhttps://lilting.ch/en/articles/qwen-qdrant-stackchan-memory-design/https://lilting.ch/en/articles/qwen-qdrant-stackchan-memory-design/A pre-implementation design for adding Qwen3-Embedding and Qdrant memory to an RTX 3050 Ti and CoreS3 voice-chat stack without depending on StackChanWorld's API. It keeps raw logs, searchable memory, persona, and body separate.Sun, 06 Sep 2026 11:17:25 GMTQwenAI AgentsM5StackRAGTTSSTTPythonhttps://lilting.ch/en/articles/modelscope-api-inference-magicube-probe/https://lilting.ch/en/articles/modelscope-api-inference-magicube-probe/My ModelScope account switched from a monthly API-Inference quota to Magicube coins, so I measured what each call costs. Qwen-Image, Z-Image, Krea-2-Turbo and FLUX worked even though none of them appeared in the /v1/models list I got; of the embedding IDs I tried only Qwen3-Embedding went through; audio paths were 404 and Wan produced no video. In these tests 400s and 404s cost nothing, while a 200 with an empty body was charged.Fri, 04 Sep 2026 18:05:00 GMTQwenModelScopeImage GenerationAPIExperimenthttps://lilting.ch/en/articles/rsa-260-factored-how-computed/https://lilting.ch/en/articles/rsa-260-factored-how-computed/862-bit RSA-260 fell on September 3, 2026 after 35 years. I verified the posted factor in Python on an M4 Mac mini, traced the viral 'seven months by hand' story to a joke, and sized the compute from RSA-250's 2,700 core-years with the GNFS formula.Fri, 04 Sep 2026 05:49:24 GMTMathCryptographySecurityhttps://lilting.ch/en/articles/lz-248kev-event-five-followup-papers/https://lilting.ch/en/articles/lz-248kev-event-five-followup-papers/One 248 keV nuclear recoil in 2.84 tonne-years, global 2.6σ (0.5% background probability), and five arXiv papers within two hours, three landing on a roughly 1 to 1.1 TeV higgsino. What LZ ruled in and out, from the preprint.Thu, 03 Sep 2026 05:50:12 GMTResearchPhysicshttps://lilting.ch/en/articles/feynman-path-integral-postulates-photon-test/https://lilting.ch/en/articles/feynman-path-integral-postulates-photon-test/Science Advances 2026: 1,419,857 single-photon paths reconstructed to test Feynman's 1948 postulates directly. Explained with arrows from energy conservation and the double slit, down to what 94% fidelity means.Thu, 03 Sep 2026 04:15:07 GMTResearchPhysicshttps://lilting.ch/en/articles/emergent-symbolic-structure-tpr-discover/https://lilting.ch/en/articles/emergent-symbolic-structure-tpr-discover/McCoy, Soulos, Linzen and Smolensky swap every GPT-OSS input-token hidden state for a closed-form tensor product formula; accuracy drops at most 2.36 points and 31 causal interventions average 0.903.Wed, 02 Sep 2026 04:20:00 GMTAILLMOpenAIResearchInterpretabilityhttps://lilting.ch/en/articles/large-emotional-world-model-lewm/https://lilting.ch/en/articles/large-emotional-world-model-lewm/LEWM predicts a 7-class emotion label for the human on screen, not a state of its own. Its 45.72% is a cosine-similarity delta vs WorldGPT, and 'self-aware' never appears in the paper.Tue, 01 Sep 2026 13:14:36 GMTAILLMResearchMultimodalhttps://lilting.ch/en/articles/cohttp-patch-probe-oss-embargo/https://lilting.ch/en/articles/cohttp-patch-probe-oss-embargo/Probes for percent-encoded traversal hit the cohttp maintainer's logs 10 minutes after the fix PR opened. Mandiant now puts mean time-to-exploit at -7 days, VulnCheck's data disagrees.Sun, 30 Aug 2026 16:57:00 GMTSecurityVulnerabilityOSSAI AgentsCVEProject GlasswingAnthropichttps://lilting.ch/en/articles/lunar-reference-time-tcl-cgpm/https://lilting.ch/en/articles/lunar-reference-time-tcl-cgpm/Draft Resolution D reaches the 28th CGPM in October alongside the leap-second vote. One lunar time scale, unscaled TCL, and why 56 and 58.7 µs/day are both correct.Sun, 30 Aug 2026 14:49:55 GMTSpaceNASAhttps://lilting.ch/en/articles/anthropic-model-hardware-standard-mhs/https://lilting.ch/en/articles/anthropic-model-hardware-standard-mhs/MHS is Anthropic's driver standard for agents running lab hardware. What it standardizes, where MCP sits underneath, and the numbers QuEra, CMU and Genentech reported.Sat, 29 Aug 2026 13:26:00 GMTMCPAnthropicClaudeAIAI Agentshttps://lilting.ch/en/articles/qwen38-flash-next-evo-x2-rocm-test/https://lilting.ch/en/articles/qwen38-flash-next-evo-x2-rocm-test/Tested on EVO-X2 (gfx1151) under Windows + ROCm: the official b10666 binary runs it at pp512 159.25 / tg128 23.26, and the earlier crash was my own workaround flag.Fri, 28 Aug 2026 13:40:00 GMTLLMLocal LLMQwenllama.cppAMDROCmExperimenthttps://lilting.ch/en/articles/qwen-capybara-mascot-timeline/https://lilting.ch/en/articles/qwen-capybara-mascot-timeline/Traced across 31 blog posts and 464 Hugging Face repos: Qwen's capybara debuted Oct 11, 2024, hit blog banners in Jan 2025, and its own edit prompts all say 'the bear'.Thu, 27 Aug 2026 10:27:00 GMTQwenAIHuggingFaceGitHubAnthropichttps://lilting.ch/en/articles/codex-suspended-tty-input-mcp/https://lilting.ch/en/articles/codex-suspended-tty-input-mcp/On an M4 Mac mini (Codex 0.150.1), stdio MCP servers keep the controlling terminal, so a child reading the tty SIGTTIN-stops the CLI. Shell tools got setsid; MCP didn't.Thu, 27 Aug 2026 06:58:00 GMTCodexOpenAIMCPmacOSDockerTerminalExperimenthttps://lilting.ch/en/articles/qwen38-flash-next-llamacpp-m1max-test/https://lilting.ch/en/articles/qwen38-flash-next-llamacpp-m1max-test/Tested on M1 Max 64GB: AtomicChat's M64 GGUF keeps the 51B N-gram table in its own shard, so a llama.cpp PR #27742 build leaves it on SSD and runs the 125B MoE at 17.6 tok/s.Thu, 27 Aug 2026 03:05:00 GMTLLMLocal LLMQwenllama.cppApple SiliconExperimenthttps://lilting.ch/en/articles/kana-icon-pose-anima-lora/https://lilting.ch/en/articles/kana-icon-pose-anima-lora/Tested on M1 Max 64GB, ComfyUI v0.30.1: i2i and Qwen-Image-Edit lost the character, so I rebuilt the pose one prompt at a time, plus why negatives do nothing at cfg 1.0.Wed, 26 Aug 2026 10:26:26 GMTLoRAAIImage GenerationAnimaComfyUIExperimenthttps://lilting.ch/en/articles/leap-second-abolition/https://lilting.ch/en/articles/leap-second-abolition/CGPM votes this October on Draft Resolution C, which stops leap seconds on May 20, 2027, driven by a 30% risk of a negative leap second by 2035. From the 2012 Reddit and 2017 Cloudflare outages to the leap second list still shipping in macOS tzdata.Tue, 25 Aug 2026 09:40:00 GMTLinuxCloudflareGohttps://lilting.ch/en/articles/anima-3-8b-qwen35-prompt-adherence/https://lilting.ch/en/articles/anima-3-8b-qwen35-prompt-adherence/Tested on M1 Max 64GB, ComfyUI v0.33.3: from-behind and a back-row drummer now pass, out-of-frame crops get worse, and Anima-Base LoRAs need a 52-block key remap.Sun, 23 Aug 2026 20:09:00 GMTLoRAAIImage GenerationAnimaAnima-BaseComfyUIQwenExperimenthttps://lilting.ch/en/articles/llama-pro-block-expansion-qwen3-blog-corpus/https://lilting.ch/en/articles/llama-pro-block-expansion-qwen3-blog-corpus/Tested on M1 Max 64GB with MLX: LLaMA Pro-style block expansion on Qwen3-0.6B-Base vs LoRA vs full FT, trained on 722 blog posts. Held-out PPL drops 14.7 to 10.5, but Wikipedia-ja rises 13.5 to 17.9, more forgetting than full FT. A 20% Wikipedia mix nearly removes it, and renumbering 28-layer LoRA keys reproduces the Anima-2.9B result in numbers.Sun, 23 Aug 2026 08:06:00 GMTLLMMLXQwenAnimaApple SiliconLoRAAIExperimenthttps://lilting.ch/en/articles/qwen-mm-plugins-core-claude-code-codex/https://lilting.ch/en/articles/qwen-mm-plugins-core-claude-code-codex/16GB M4 Mac mini test of QwenLM's Qwen-MM-Plugins core. Claude Code adds a plugin_ prefix that breaks --allowedTools, codex exec rejects MCP calls by default. PDF, STL and video read with no API key.Sat, 22 Aug 2026 05:20:00 GMTQwenMCPClaude CodeCodexAI AgentsMultimodalExperimenthttps://lilting.ch/en/articles/leap-year-magic-number-z3/https://lilting.ch/en/articles/leap-year-magic-number-z3/A three-instruction leap-year test hides the constants 1073750999, 3221352463, and 126976. I rebuilt an 18-bit version with Z3, then traced how the multiply, mask, and threshold encode divisibility by 4, 100, and 400.Sat, 22 Aug 2026 05:11:50 GMTCAlgorithmsMathExperimenthttps://lilting.ch/en/articles/anima-2-9b-layer-expansion-lora-compat/https://lilting.ch/en/articles/anima-2-9b-layer-expansion-lora-compat/Tested on M1 Max 64GB, ComfyUI v0.33.3. Anima-2.9B inserts 12 DiT blocks, so Anima-Base LoRA keys hit the wrong layers with zero warnings, 0/3. Renumbering the keys gives 3/3.Sat, 22 Aug 2026 04:50:00 GMTLoRAAIImage GenerationAnimaAnima-BaseComfyUIQwenExperimenthttps://lilting.ch/en/articles/esprimo-g5010-fujitsu-logo-usb-freeze/https://lilting.ch/en/articles/esprimo-g5010-fujitsu-logo-usb-freeze/My Fujitsu ESPRIMO G5010/E froze at the FUJITSU logo with any bootable USB inserted. Disabling the boot logo in BIOS fixed it. D3804-A1x BIOS changelog included.Thu, 20 Aug 2026 17:15:00 GMTLinuxUbuntuUEFIhttps://lilting.ch/en/articles/fast-day-of-week-mersenne-benchmark/https://lilting.ch/en/articles/fast-day-of-week-mersenne-benchmark/Benchmarked on an M4 Mac mini: Ben Joffe's 2-instruction weekday hack beats plain %7 by 1.6-6.4x in clang, Rust, and V8, loses 3x in CPython. Plus a 25x V8 -0 deopt trap.Thu, 20 Aug 2026 09:10:00 GMTCRustJavaScriptPythonAlgorithmsBenchmarkApple SiliconMathExperimenthttps://lilting.ch/en/articles/qwen38-27b-evo-x2-rocm-mtp-test/https://lilting.ch/en/articles/qwen38-27b-evo-x2-rocm-mtp-test/Tested on GMKtec EVO-X2 (Ryzen AI Max+ 395): Q8_0 + MTP beats a 4-bit M1 Max at 22 tok/s, thinking burns 32,712 chars before any HTML, and the NSFW refusal line moves.Wed, 19 Aug 2026 07:41:00 GMTLLMLocal LLMQwenllama.cppAMDROCmExperimenthttps://lilting.ch/en/articles/qwen38-27b-mlx-ollama-m1max-test/https://lilting.ch/en/articles/qwen38-27b-mlx-ollama-m1max-test/Tested on M1 Max 64GB: Qwen3.8-27B hits ~19 tok/s on both MLX and Ollama, but the default reasoning_effort=xhigh blew thinking up to 50,373 chars. Why Ollama dodges it.Tue, 18 Aug 2026 07:00:04 GMTLLMLocal LLMQwenOllamaMLXApple SiliconExperimenthttps://lilting.ch/en/articles/sqlite-fake-cve-advisory-database-unreviewed/https://lilting.ch/en/articles/sqlite-fake-cve-advisory-database-unreviewed/Checked August 17, 2026 via the GitHub Advisory Database API: all six fake SQLite CVEs rejected by MITRE still show CVSS up to 9.8 as type unreviewed, withdrawn_at null.Mon, 17 Aug 2026 04:28:34 GMTSecurityCVEVulnerabilitySQLiteAIOSSGitHub