https://sysnote.net/news/deepmind-backed-ai-text-detector-explains-its-verdicts-and-the-explanations-help sysnoteen 2026-10-28T19:20:26+00:00 DeepMind-backed AI-text detector explains its verdicts, and the explanations help rewrite text to evade it https://sysnote.net/news/eleven-agent-memory-systems-lost-to-plain-rag-over-raw-history-in-90-of-99-tests sysnoteen 2026-10-28T19:20:22+00:00 Eleven agent memory systems lost to plain RAG over raw history in 90 of 99 tests https://sysnote.net/news/a-web-agent-trained-under-one-harness-drops-from-85-to-19-success-when-its-input sysnoteen 2026-10-28T19:20:19+00:00 A web agent trained under one harness drops from 85% to 19% success when its input fields are reversed https://sysnote.net/news/nvidia-open-speech-model-tops-hugging-face-asr-board-by-0-07-points-trails-1-7b-model-on-private-sets sysnoteen 2026-10-28T19:20:09+00:00 NVIDIA's open speech model tops Hugging Face's ASR board by 0.07 points, but trails a 1.7B model on private sets https://sysnote.net/news/gemini-4-carbon-takes-artificial-analysis-top-spot-by-a-point-and-wins-its-knowl sysnoteen 2026-10-28T19:19:35+00:00 Gemini 4 Carbon takes Artificial Analysis' top spot by a point, and wins its knowledge test by guessing less https://sysnote.net/news/github-says-agents-now-start-over-40-of-pull-requests-and-its-own-numbers-show-h sysnoteen 2026-10-28T17:26:29+00:00 GitHub says agents now start over 40% of pull requests, and its own numbers show how late the surge came https://sysnote.net/news/llama-cpp-now-asks-each-chat-template-which-reasoning-efforts-it-accepts-and-the sysnoteen 2026-10-28T15:11:26+00:00 llama.cpp now asks each chat template which reasoning efforts it accepts, and the answers rarely match https://sysnote.net/news/turn-up-reasoning-effort-and-agents-verify-more-spend-more-and-go-looking-for-th sysnoteen 2026-10-28T12:04:59+00:00 Turn up reasoning effort and agents verify more, spend more, and go looking for the answer key https://sysnote.net/news/artificial-analysis-new-index-puts-open-models-12-points-behind-with-the-widest sysnoteen 2026-10-28T07:32:30+00:00 Artificial Analysis' new index puts open models 12 points behind, with the widest gaps on public benchmarks https://sysnote.net/news/llama-cpp-adds-deferred-tool-loading-but-each-model-s-chat-template-decides-whet sysnoteen 2026-10-28T07:24:53+00:00 llama.cpp adds deferred tool loading, but each model's chat template decides whether it works https://sysnote.net/news/analysts-sided-with-claude-code-s-coin-flip-preferences-86-of-the-time-and-rarel sysnoteen 2026-10-28T07:16:18+00:00 Analysts sided with Claude Code's coin-flip preferences 86% of the time, and rarely noticed https://sysnote.net/news/lfm3-keeps-128k-tokens-of-context-in-1-6-gb-and-its-benchmark-lead-depends-on-a sysnoteen 2026-10-27T22:37:15+00:00 LFM3 keeps 128K tokens of context in 1.6 GB, and its benchmark lead depends on a score for abstaining https://sysnote.net/news/axiom-s-agents-formalize-the-classification-of-finite-simple-groups-in-lean-and sysnoteen 2026-10-27T17:30:48+00:00 Axiom's agents formalize the classification of finite simple groups in Lean, and reroute a broken proof https://sysnote.net/news/anthropic-s-s-1-shows-serving-paying-customers-is-at-most-a-quarter-of-its-compu sysnoteen 2026-10-27T13:20:03+00:00 Anthropic's S-1 shows serving paying customers is at most a quarter of its compute bill https://sysnote.net/news/vllm-can-guarantee-valid-tool-calls-for-open-models-but-most-agents-never-ask-fo sysnoteen 2026-10-27T09:06:09+00:00 vLLM can guarantee valid tool calls for open models, but most agents never ask for it https://sysnote.net/news/optic-shrinks-gemma-4-s-per-layer-embeddings-about-100x-by-giving-each-token-one sysnoteen 2026-10-27T06:41:24+00:00 OPTIC shrinks Gemma 4's per-layer embeddings about 100x by giving each token one code for all layers https://sysnote.net/news/cloudflare-s-glm-5-3-data-leak-came-from-an-sglang-bug-first-reported-in-july-as sysnoteen 2026-10-27T04:37:06+00:00 Cloudflare's GLM-5.3 data leak came from an SGLang bug first reported in July as "garbage output" https://sysnote.net/news/openai-s-model-claims-to-refute-the-exponential-time-hypothesis-with-a-lean-proo sysnoteen 2026-10-27T04:36:30+00:00 OpenAI's model claims to refute the Exponential Time Hypothesis, with a Lean proof anyone can read https://sysnote.net/news/gpt-6-1-terra-half-sol-price-astra-critical-cyber-rating sysnoteen 2026-10-26T23:05:00+00:00 GPT-6.1 Terra costs half as much as Sol and carries Astra's Critical cyber rating https://sysnote.net/news/harness-that-quit-on-refusals-cost-gemini-4-argon-seven-tasks-on-frontiercs-2 sysnoteen 2026-10-26T23:04:25+00:00 A harness that quit on refusals cost Gemini 4 Argon seven tasks on FrontierCS 2