https://sysnote.net/news/anthropic-s-flashprove-zero-knowledge-proofs-for-llama-4-maverick-at-110-tokens sysnoteen 2026-11-03T21:03:53+00:00 Anthropic's FlashProve: zero-knowledge proofs for Llama 4 Maverick at 110 tokens/s, and a strict check stock engines fail https://sysnote.net/news/gpt-6-astra-s-qwen-weight-edit-holds-up-in-a-rerun-at-51-0-on-math-500-honnibal sysnoteen 2026-11-03T17:08:22+00:00 GPT-6 Astra's Qwen weight edit holds up in a rerun at 51.0% on MATH-500; Honnibal and Goedecke say the "natively" framing doesn't https://sysnote.net/news/revolut-runs-glm-5-3-for-2-000-engineers-on-56-b200s-what-david-abitbol-s-talk-a sysnoteen 2026-11-03T14:32:27+00:00 Revolut runs GLM 5.3 for 2,000 engineers on 56 B200s: what David Abitbol's talk actually says https://sysnote.net/news/prefrontal-s-latent-sparse-attention-speeds-up-one-attention-layer-72x-at-10m-to sysnoteen 2026-11-03T09:46:50+00:00 Prefrontal's Latent Sparse Attention speeds up one attention layer 72x at 10M tokens; end to end, the gain is 9x https://sysnote.net/news/hugging-face-s-carbon-x-an-open-14b-model-that-reads-raw-dna-and-answers-in-engl sysnoteen 2026-11-03T09:10:31+00:00 Hugging Face's Carbon-X: an open 14B model that reads raw DNA and answers in English, close to GPT-6.1 Sol on its own DNA benchmarks https://sysnote.net/news/stanford-re-tests-50-famous-llm-findings-on-20-models-44-have-expired-and-the-on sysnoteen 2026-11-03T08:39:55+00:00 Stanford re-tests 50 famous LLM findings on 20 models: 44% have expired, and the ones that last are about sycophancy and self-preference https://sysnote.net/news/openai-s-pando-a-git-backend-that-drops-repo-wide-consistency-so-codex-agents-ca sysnoteen 2026-11-03T08:28:51+00:00 OpenAI's Pando: a Git backend that drops repo-wide consistency so Codex agents can push in parallel, cutting median push latency about 5x https://sysnote.net/news/gemini-2-5-pro-calls-will-be-answered-by-gemini-3-8-flash-from-december-1-and-2 sysnoteen 2026-11-03T08:28:21+00:00 Gemini 2.5 Pro calls will be answered by Gemini 3.8 Flash from December 1, and 2.5 Flash by 3.5 Flash-Lite https://sysnote.net/news/claude-sonnet-5-gpt-6-1-sol-and-grok-4-7-each-did-mouse-brain-surgery-at-exin-th sysnoteen 2026-11-03T08:16:26+00:00 Claude Sonnet 5, GPT-6.1 Sol and Grok 4.7 each did mouse brain surgery at Exin Therapeutics, and picked three different gene therapies https://sysnote.net/news/mit-team-says-gpt-6-astra-can-read-and-write-neural-network-weights-its-6x-qwen sysnoteen 2026-11-03T08:16:14+00:00 MIT team says GPT-6 Astra can read and write neural network weights; its 6x Qwen math gain comes from a single 50-pass run https://sysnote.net/news/claude-s-messages-api-gets-message-threads-in-beta-send-only-the-new-turn-not-th sysnoteen 2026-11-02T21:15:08+00:00 Claude's Messages API gets message threads in beta: send only the new turn, not the whole conversation https://sysnote.net/news/gpt-rosalind-6-uses-gpt-6-astra-s-weights-with-no-extra-training-its-system-card sysnoteen 2026-11-02T18:36:54+00:00 GPT-Rosalind-6 uses GPT-6 Astra's weights with no extra training, its system card addendum says https://sysnote.net/news/anthropic-s-s-1-a-puts-its-compute-commitments-at-895-4-billion-serving-customer sysnoteen 2026-11-02T18:36:50+00:00 Anthropic's S-1/A puts its compute commitments at $895.4 billion; serving customers still takes about a quarter of the compute bill https://sysnote.net/news/semianalysis-gdpx-benchmark-tops-out-at-12-for-claude-opus-5-5-and-safety-classi sysnoteen 2026-11-02T14:30:17+00:00 SemiAnalysis' GDPx benchmark tops out at 12% for Claude Opus 5.5, and safety classifiers shape the ranking https://sysnote.net/news/openai-now-limits-how-many-paid-chatgpt-accounts-one-device-or-card-can-use-clos sysnoteen 2026-11-02T14:30:12+00:00 OpenAI now limits how many paid ChatGPT accounts one device or card can use, closing the multi-subscription route to more Codex usage