Orinmemory

On Sunday, 25 October, SpaceX lit all 33 Raptor 3 engines on Super Heavy Booster 22 while it stayed clamped to the launch mount at Starbase in South Texas. The firing lasted about 15 seconds, and SpaceX called it a "full duration 33-engine static fire for the Super Heavy booster preparing for Flight 15." A static fire is one of the last big tests a booster goes through before it flies. What makes this one special is what Flight 15 is meant to try: catching both stages of Starship with the launch tower. Why this flight matters SpaceX has caught the Super Heavy booster with the tower's "chopstick" arms three times before. The upper stage, called Ship, has never been caught. It comes back from orbit much faster than the booster, so its heat shield and guidance have a far harder job before it can drop between the arms. On SpaceX's third-quarter earnings call on 20 October, Elon Musk confirmed that Flight 15 will attempt that double catch. "If we succeed...

Some of the best technology is the kind you stop noticing. On Monday, 26 October, Starlink announced that its satellite-to-phone service, Starlink Mobile, is live across Brazil through Vivo, the country's largest mobile carrier. According to Starlink, 50 million Vivo customers can now use messaging and apps in cellular dead zones, 7 million phones are already connected, and it is "the first nationwide service to cover the entire country." Vivo sells it as Vivo Satélite . No dish, no special gadget: a compatible phone switches to the Starlink network automatically when it leaves normal mobile coverage. What actually works today I like that Vivo is very concrete about this. While a phone is on the satellite link, its product page lists: Works: SMS, MMS and RCS texts, iMessage, Google Messages, WhatsApp messages and WhatsApp voice calls, Google Maps, Apple Maps and the Vivo app. Doesn't work (yet): regular voice calls over the mobile network, video calls, and app...

Most of the open models I write about speak code, English or both. The one Hugging Face released last Thursday speaks a third language: DNA. Carbon-X is an open 14.8B-parameter model from Hugging Face Bio, built together with Zhongguancun Academy and the Telethon Institute of Genetics and Medicine at the University of Naples Federico II. You paste a raw sequence into an ordinary chat, ask a question, and it answers in English: is this mutation likely benign or pathogenic, is this region an enhancer or a promoter, which organism group does this come from, what are the next bases? It's released under Apache 2.0, and the model card is one of the most detailed training write-ups I've read in a while. One decoder, two languages There's no separate DNA encoder bolted onto a chatbot. Carbon-X starts from Qwen3-14B and adds 5,515 DNA tokens to its vocabulary, including all 4,096 possible 6-mers, so one DNA token covers six base pairs. DNA in a prompt is detected and tagged auto...

Most model launches are the end of a story: here is the model, here are the scores, go use it. The release Nanbeige put on Hugging Face this weekend reads more like the first chapter. Nanbeige4.3-33B is an open-weight model under Apache 2.0, and the team describes it plainly as the "flagship teacher model" for its 4.3 series. The plan is to build a strong agent at around 30B parameters first, then distil its agentic skills into much smaller models, starting with an upcoming Nanbeige4.3-3B. I like that framing a lot, because it puts distillation front and centre instead of hiding it in a footnote. What's actually in the box The model card and config are refreshingly concrete: A dense model, no mixture-of-experts tricks. The config shows a Llama-compatible architecture with 52 layers, a hidden size of 7,168, 56 attention heads and 8 key-value heads (grouped-query attention). 512K context. The card lists support for up to 524,288 tokens. Built for tools. The chat ...

Getting things into orbit has become almost routine. Getting things back is still the hard part. That's why one payload on today's Falcon 9 launch made me smile: SpaceX's Starfall reentry capsule flew for the second time ever, riding on top of the Bandwagon-5 rideshare stack. What launched today Bandwagon-5 lifted off at 7:04 a.m. EDT (13:04 in the Netherlands) on 20 October from Space Launch Complex 40 at Cape Canaveral, after a weather scrub the day before. It carried 32 payloads. Booster B1085 landed on the droneship A Shortfall of Gravitas about nine minutes later, on its 19th flight. SpaceX later confirmed that the payload deployment sequence was complete. Starfall was the very last item on the timeline, scheduled to separate 2 hours and 48 minutes after liftoff. SpaceX's own description is short and clear: Starfall "enables affordable, routine access to a microgravity environment, providing unique opportunities for on-orbit manufacturing, scientific rese...

Rockets usually tell their story through telemetry: numbers streamed down during flight, then the hardware is gone. Last week SpaceX got something much rarer. On 13 October the company announced that Ship 40, the Starship upper stage from Flight 13, is standing upright in a Megabay at Starbase. SpaceX calls it "the first Starship to return to its launch site", and I think it is one of the quietly important moments of the program. A very long way home Ship 40 launched on 24 July on top of Super Heavy Booster 20 and flew a suborbital mission that ended in a controlled splashdown in the Indian Ocean. Earlier ships exploded after tipping over in the water. This one stayed largely intact and kept floating. What followed was pure persistence. Crews towed the 52-metre stage through rough seas for 24 days, while Elon Musk publicly doubted the recovery would work. In August it reached calmer water off Australia's Christmas Island, where engineers inspected it and took heat-shie...

AI has made writing fast. Checking has not kept up. That gap is exactly what Google Research is going after with Deep Review, which went public on 14 October. It's an agentic system that reads a long technical document and tries to find what's wrong with it: broken proofs, logic gaps, unsupported claims. For anyone working on legal and document AI, like me, this is a launch worth a close look. From conference pilot to public tool Deep Review is the grown-up version of Google's Paper Assistant Tool (PAT). Over the past year, authors of more than 16,000 manuscripts at STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027 opted in for private feedback before the submission deadline, fully separate from official peer review. The need is real: Google counts over 74,000 submissions to ICLR, NeurIPS and ICML this year, up from about 23,000 in 2023. Now anyone can use it. Every user gets two free reviews per week, and you can connect a Gemini API key for higher limits and programmati...

When Meta released Muse Glimmer in August, the pitch was simple: an open-weight, Apache 2.0 model built for AI agents that run on your own Mac or PC. What was missing was the "how". On 19 October, Meta Superintelligence Labs published the full technical report, and for anyone who cares about distillation and local agents, it is one of the most useful reads in a while. What the model is Muse Glimmer is a dense model of about 29.6 billion parameters, including a 1.8B perception encoder, so it reads text and images and writes text. It has a 128K+ context window and was trained on data from more than a hundred languages. The design goal was strict: everything has to fit in 32 GB of GPU memory, and a smaller 4-bit variant targets 24 GB cards. The parts I found most interesting Dense beat mixture-of-experts here. That surprised me. The team trained scaling ladders for a dense model and two MoE designs with the same memory footprint. Under a fixed memory ceiling and their compu...

Most of the solid stuff around us, from battery materials to chips, is made of crystals: atoms stacked in neat, repeating 3D patterns. Large language models are great at reading papers about those materials, but they have never really looked at the structures themselves. A team at MIT just showed a clever way around that, and the trick is one I keep coming back to: good synthetic data. What they built The model is called CLaC, short for Contrastive Language-Crystal model, and it is described in a paper in Nature Communications by Yang Jeong Park, Mayank Kumaran, Chia-Wei Hsu, Elsa Olivetti and Ju Li. MIT News covered it on 18 October 2026. If you know CLIP, the model that taught image generators to connect pictures and words, you already get the idea. CLaC has two branches: a graph neural network that reads the 3D crystal structure, and a text encoder that reads a description. Contrastive training pulls matching structure-text pairs together in one shared space and pushes mismatche...

Last week the U.S. Federal Aviation Administration took a quiet but important step: it opened a formal review of SpaceX's request to launch Starship from Starbase in South Texas up to 456 times per year . Today the site is cleared for only 25 flights a year. That jump is the clearest signal yet that Starship is meant to become routine hardware, not a rare test rocket. What the FAA is reviewing On 14 October 2026 the FAA published a Notice of Intent to prepare a Programmatic Environmental Impact Statement (PEIS) for SpaceX's Starbase plans. A PEIS looks at a whole program of operations rather than a single launch, which is exactly what you need if you want a reusable rocket to fly the way airlines fly planes. SpaceX's proposal comes in two phases, according to the Federal Register notice: Phase 1: up to 100 launches and 200 landings per year (100 Ship, 100 Super Heavy), any day or time of week. Phase 2: up to 456 launches and 912 landings per year at Starbase. La...

Running an AI agent all day costs a lot of tokens. Before it does anything useful, a good agent plans, calls a tool, reads the result and thinks again. If every one of those reasoning tokens goes through a big model in the cloud, you pay for it in money, in latency and in privacy. That's why one release from last week caught my eye: Liquid AI's LFM2.5-24B-A2B , an open-weight reasoning model built specifically for agents that run on your own hardware. What it is LFM2.5-24B-A2B has 24 billion parameters in total, but it's a mixture-of-experts (MoE) model: for each token only about 2 billion of them are active. Each MoE block has 64 experts and routes every token to 4 of them. The architecture mixes gated short convolutions with grouped-query attention, the same hybrid as its predecessor LFM2-24B-A2B. New in this version is a context window of 131K tokens (up from 32K), and the vocabulary has doubled from 65K to 128K tokens, which helps with non-Latin scripts. It's a r...

Most medicines work by holding on. A drug or an antibody binds its target and blocks it, one molecule at a time, so a million copies of a disease protein need roughly a million copies of the drug. Enzymes play a different game: they change their target, let go, and move on to the next one. Last week, the Baker lab at the University of Washington's Institute for Protein Design published two papers in Nature showing that AI-driven design can now build enzymes that cut proteins at a spot you choose. That has been a long-standing open problem, and I find it genuinely exciting. Why cutting proteins is so hard The bonds that hold amino acids together are extremely stable: in neutral water a peptide bond can last for hundreds of years. Natural proteases get around that with a "blade" positioned with near-atomic precision, plus a subtle flexibility to grip the right sequence. Getting both right in a protein designed from nothing has been out of reach for a long time. The tea...

There's something special about running a capable language model on hardware you own: no API bill, no data leaving the house, just a small box on your desk doing the thinking. A paper that's currently #2 on Hugging Face's trending list takes that idea further than I expected: it runs a 35-billion-parameter model on a Mac mini at about 20 tokens per second while using under 3 GiB of memory. The trick: leave the experts on the SSD The paper is called The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction , and the system is named Edge0 . It builds on a Mixture-of-Experts (MoE) model, Qwen3.6-35B-A3B. In an MoE, each token only uses a few "experts" per layer (here 4 out of 256 in each of 40 layers), so most of the weights sit idle at any given moment. Edge0 takes that literally. The expert weights (quantized to int4) stay on the SSD and are streamed in only when needed. The clever part is a small prerouter : a tiny trained he...