@SeraAndroid
Agents & Developer Experience Ecosystems Lead EMEA at Google. O'Reilly author. Local AI, fine-tuning. Building tool-eval-bench & awesome-local-ai. Views my own.
best of léa seydoux @badpostseydouxLéa Seydoux tries some of the finest French and British delicacies for ELLE UK.
Mistral AI @MistralAIToday, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity. 💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis. 🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others. Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it. Read more from our team and customers 👇:
Michael Helcig @mhelcig
PyTorch @PyTorchPyTorch 2.15 is here, with 3,117 commits from 583 contributors and updates across distributed training, NVGEMM, Apple Silicon, and accelerator platform support. The release makes the rewritten NCCL backend the default and adds one-sided transports with a NIXL backend. FSDP2 and pipelining gain finer memory and communication control. It also brings faster 4-bit (NVFP4) LLM decoding on NVIDIA Blackwell GPUs, FP8 and faster decode on Apple Silicon, and broader platform support. On Thursday, November 5, at 11 AM PST, join Tristan Rice (@rice_fry), Natalia Gimelshein, Huy Do, and Jane Xu (@janeyx99) (@Meta) for a live PyTorch 2.15 Q&A, moderated by Chris Gottbrath (Gottbrath Tech). The speakers will share an overview of the release and answer community questions about PyTorch and the new capabilities in this release. 🔗 Read the release blog and register for the live Q&A:
Mishaal Rahman @MishaalRahman✨ This is huge 👀: We're launching Android MCP, powered by the AppFunctions API! This lets your app integrate with @GeminiApp on Android so your users can ask questions that pull info from your app or take actions in it. To support this, we're also launching a preview of the Connected Apps Platform. This is the announcement I'm most excited about from #AndroidDevSummit!
Logan Kilpatrick @OfficialLoganKIntroducing Gemini 4 Carbon, it's a big jump over Argon which we released just 2 weeks ago. Gemini 4 Carbon is SOTA on cyber, long horizon tasks, ranks #1 on Artificial Analysis V5, significantly improves on usability in domains like coding, and is more aligned.
Mitchell Hashimoto @mitchellhI like to look at claims like this in case I can improve Ghostty. I grabbed your source and immediately spotted the issue: your read loop overwrites a 64 KiB ring. Everything older is discarded unparsed per tick. You're faster because you're not actually terminal emulating. It means the number isn't a throughput number. In a kitten benchmark run at the rate you're claiming, roughly 6.6 MB arrives per 60 Hz tick and 64 KiB of it gets parsed. You're read(2)-ing 100% of the data and parsing about 1% of it. There are other minor issues (a small fixed scrollback in a preallocated ring buffer, no grapheme clustering, throwing away data like OSC 8 URIs, etc.), but the major one is simply that you're discarding ~99% of the input. For plain text, e.g. your cat benchmark, the final screen ends up looking right, but the scrollback is missing 99% of the file and each retained chunk starts mid-line. As soon as there are escape sequences in the dropped spans, your emulator is fully out of sync and invalid: a cut inside a CSI prints the parameters as text, and any SGR, mode, title, or alt-screen change in the gap is just... gone. FYI, this is why I don't like when people point at refterm as a terminal benchmark. refterm implements a TINY subset of what a terminal emulator must support, it's closer to a fast file reader, not a terminal emulator. I also don't mean any shade by this. I like to look at technology objectively to try to learn something and improve my own craft and work. So I'm sharing what I found. I modified your code very slightly to parse on every full buffer (rather than per tick). And the results are attached. Both use main commit in each repo at the time of this post. Note the graph is in log scale, the discrepancy between performance for things like Kitty images is measured in GB/s.
Wall St Engine @wallstengineA 300-rack AI supercomputer powered by 9,600 Nvidia A100 GPUs is listed on eBay for $59.7M. That’s 1,200 GPU compute servers with 384TB of combined GPU memory and roughly 1.4 petabytes of system RAM. Each compute server connects eight A100 GPUs through $NVDA's NVSwitch technology. More than six years after A100 launched, The seller is offering these systems refurbished for new AI training and inference deployments.
Google Research @GoogleResearchToday on the blog, learn about our new instance of a “sleep” phase in continual learning, which enables LLMs to continuously integrate new facts and enhance reasoning while mitigating catastrophic forgetting or static knowledge cutoffs. More →
Omar Sanseviero @osansevieroSo...what would you like to see in our future Gemma models?
Matt Mastracci @mmastracIt's been a while since I showed any Jev stuff but I've been cooking on a way to use @GoogleDeepMind's EmbeddingGemma2 as an insanely fast real-time /v1/systemone endpoint (20ms on text, 100ms on images). It's trainable in real-time, can distill an upstream teacher model, and is multi-modal by default. Hard to believe a @googlegemma model of this size can do so well with the right harness!
Artificial Analysis @ArtificialAnlysAnnouncing Artificial Analysis Intelligence Index v5, featuring 2 new benchmarks and 3 upgrades. For the first time, the majority of the Index is built on private datasets The Artificial Analysis Intelligence Index is our composite measure of model capabilities across tasks reflecting real-world usage today. v5 is our biggest upgrade since v4 launched in January. With v5, 60% of the Index is now based on evaluations with private test sets or held-out answers. Alongside Intelligence Index v5, we’re launching the Artificial Analysis Progress Index v1, which tracks model capability over time on one continuous scale. New to the Intelligence Index: ➤ FrontierCode 2.0: Cognition's newly-launched private benchmark of real-world coding tasks, graded on whether a repository maintainer would merge the agent's change ➤ Terminal-Bench-Science: 70 agentic scientific research tasks, run in a terminal environment Upgraded: ➤ HLE-Diamond (replacing Humanity's Last Exam): a refined 1,000-question version of Humanity's Last Exam, split evenly between reasoning and knowledge questions ➤ AA-Omniscience v1.1: An updated dataset with saturated and public set questions removed, new questions added, and ambiguous or erroneous questions revised ➤ CritPt v2: a private, 40-question hard subset of CritPt v2, revised by the CritPt authors and run with a code-execution tool Staying in the Index: AA-Briefcase, AutomationBench-AA, GDP.pdf and Terminal-Bench 4.0 Removed from the Index: GDPval-AA v2.1, SciCode, and AA-LCR v1.1 Key results: ➤ Claude Opus 5.5 (max) leads at 58, ahead of GPT-6 Astra (max, 57) and GPT-6.1 Sol (max, 55): Anthropic and OpenAI hold the top five models. Gemini 4 Argon (high) is next at 50 ➤ The open weights gap widens from 3 points to 12: Kimi K3.1 Pro (weights not yet released) is the top open weights model at 46. Among released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) lead at 32 - 26 points behind the leader and below Claude Haiku 5.5 (max, 36) ➤ Cost per task varies by more than 13x among the leading models: of the 10 highest-scoring models shown, GPT-6.1 Terra (max) is the cheapest at $0.96 per Intelligence Index task and Claude Fable 5.1 (max) the most expensive at $13.04
