🚀 The full DeepSeek App experience, now via API Search, files, links, safety, and voice. The capabilities behind the DeepSeek App are now available through the DeepSeek API. Enable built-in features with a few parameters, or call standalone APIs such as OCR, ASR, and TTS. Build
Week 46 · November 9–15, 2026
People are talking
What researchers, labs and projects in our four lanes are posting on X, from the last three days. Hottest and newest first; promos, hiring posts, giveaways and event plugs are filtered out.
58 posts from 28 accounts · last 3 days · updated Fri Nov 13, 2026 · 9:19 AM CT
llama.cpp v1.0.0 is now available huggingface.co/blog/ggml-org/llama-cpp-v1
Meet Gemini 4 Hydrogen: built for rapid coding, multimodal synthesis, and everyday workflows. It finishes tasks 2.5x faster than 3.8 Flash and surpasses Argon on the @ArtificialAnlys Intelligence Index.
Accurate weather forecasting starts with how clearly we can observe the atmosphere in real-time. 🌍 That’s why we’re releasing PARDIG and SatWeave - the foundational datasets behind WeatherNext 3 - to help researchers and developers build new climate applications 🧵
Now in beta: Businesses in ChatGPT. Claim your business, manage your information, and get insights on how people engage with you. Eligible businesses can also build a Business Agent to answer customers’ questions and help them explore what you offer. openai.com/index/introducing-businesses-…
Qwen4 is coming. Deeper knowledge. Richer world understanding. Stronger coding and agentic capabilities. Built for real-world tasks — from complex coding and knowledge work to front-end builds and professional software workflows. Qwen4-Max is now in preview on Alibaba’s Token
Atlas is now available to everyone!! I’m very excited to see what users can build with our newest world model! 🤩🌎
We’re opening up access to Atlas to everyone today. We’ve built apps and API endpoints to help you use Atlas any way you can imagine: bring your ideas for building a 3D world and come away with keyframes, video clips, or 3D Gaussian splats. https://t.co/Y9vhcrwbWu
Just received this in the mail today: The Welch Labs - Illustrated Guide to AI". What an astonishing, gorgeous work of art! Bravo, Stephen Welch and team!
Your heart, brain, and immune system share the same DNA sequence. So how can a single-letter change affect disease in some cells, while sparing the rest? AlphaGenome 2 is our new AI model that zooms in on specific cells to help scientists understand what drives disease. 🧵
For the first time, AI has mathematically proven @signalapp’s protocol is secure down to the code level. Using our breakthrough AlphaProof Code framework, we verified the messaging app’s cryptographic system – which protects billions of private messages every day.
Hoping Nemotron keeps going. Thankfully, open weight LLMs have some alternatives. But their open data and full training recipes would be much harder to replace.
We optimized DeepSeek-V4.1-Flash on vLLM for Vera Rubin NVL72 and reached 58× the throughput of GB300 NVL72 at 1.7 million tokens/s/GPU and 387 tokens/s/user on SemiAnalysis AgentX. Across the curve: 🔹 2.52 million tokens/s/GPU at peak throughput (2.7× GB300) 🔹 1.7 million
Sending support to all my Nvidia friends today, especially at Nemotron and HuggingFace. I hope your work comes back even stronger after a transition period. It’s absolutely crucial.
Can robots approach human-speed manipulation? Teleop yields slower data, while UMI tracking degrades with higher speeds. Flash ALOHA has people directly guide the robot, yielding fast data with accurate tracking. Paper, code, hardware: johnyao93.github.io/flash-aloha
Introducing ⚡️Flash ALOHA: Learning Fast and Precise Manipulation! With just 50 demonstrations, Flash ALOHA learns to autonomously complete dynamic and precise tasks at near human speed. Everything open sourced! (1/11) Team: @Qi_Wu577 @YihuaiGao @chelseabfinn @zipengfu https://t.
Run big policies on a remote GPU ⚡🤖 Your robot keeps moving while the GPU thinks. New in LeRobot 0.7: the robot computer runs cameras, motors and recording, and a policy server on a GPU machine predicts the next chunk of actions. An action buffer asks for the next chunk
Conversational AI is going to transform education and deliver personalized tutoring at scale. But we have to build it right. Most current AI tools in education optimize for the wrong thing: getting the answer. But the point of homework is not the output, it's learning to think
We’ve raised $51 million to rethink how kids learn with AI. Our bet: conversational learning. Language isn’t just how we communicate. It’s how we think. That’s the idea behind Gakku, our AI learning coach. The learning process itself is the product—not something to automate away.
Writing about math for a wide audience is tricky. Here are some pitfalls we fell into, and lessons we learned along the way. ams.org/journals/notices/202611/noti3425…
LLM training is hard. Hard to be a jack of all trades.
im surprised there isnt more pretrain-postrain codesign i guess companies really do ship their org charts
Super fun week @corl_conf meeting the brilliant next generation of robotics experts!
Introducing GR00T 2: our most advanced, open System One model for robotics. It runs entirely on your robot’s onboard compute, fully untethered, and you own every weight of it. GR00T 2 features: 1. Long-horizon tasks: dexterous manipulation and whole-body control chained over
We're excited to announce the 2026 recipients of the Google PhD Fellowship! 148 new PhD students across 38 countries, with $13M in support from @Googleorg. Their work spans quantum theory, language model reasoning, medical imaging and more: goo.gle/phd-fellows-2026
“Go Turns 17” by Austin Clements, for the Go team — go.dev/blog/17years #golang
Write Minecraft mods in Swift. Connect to robots running ROS 2. Share code across iOS and Android. See what's new in Swift this month in our latest digest: swift.org/blog/whats-new-in-swift-novemb…
Autonomous, uncut, 1x speed. This is GR00T 2 controlling real dexterous hands to perform a chemistry lab experiment (and clean up after itself) end to end. Episode 1 of a new series! I’ll keep dropping real-time robot videos, with none of the “4x” or “8x” labels you usually see
Introducing GR00T 2: our most advanced, open System One model for robotics. It runs entirely on your robot’s onboard compute, fully untethered, and you own every weight of it. GR00T 2 features: 1. Long-horizon tasks: dexterous manipulation and whole-body control chained over http
Today on the blog we introduce an investigational ML framework that translates passive wearable signals into a modifiable coachable health age, uses an action-conditioned world model to simulate personalized 12-week "what-if" habit trajectories, & expands to routine blood
🎉 Day-0 support for @thinkymachines’ Inkling-2-Small in vLLM! A 500B MoE with text + image inputs and text output. Bring the image into the conversation: ask a question with visual context, and get a text response from the same model. Under the hood: a new ViT image encoder,
Inkling 2 Small’s weights are now available for download on HuggingFace under an Apache 2.0 license. Make Inkling your own! https://t.co/KS5zd97XSd
SteerDistill Quantizing LLM Agents via Closed-Loop Teacher Steering paper: huggingface.co/papers/2611.13410
Thank you Junfan, @tianyurobot, @letian_fu, all speakers, attendees, and organizers of the first Workshop on AR — it won’t be the last! agentic-robotics-workshop.github.io
CoRL Agentic Robotics Workshop was absolutely FULL. Thank you all for making our @corl_conf Agentic Robotics Workshop one of the hottest rooms at CoRL 2026! Agentic Robotics is taking off. 🤖🚀 @Ken_Goldberg @GuanyaShi @HaoruXue https://t.co/Vqk5MH9FfX
In reasoning model development, it's always been this interesting tension between training/optimizing chain-of-thoughts (CoTs) vs not touching them to preserve monitorability. CoTs are cool because we can check what the model is up to (e.g., whether it cheats or not) as long as
(1/10) Today, we’re sharing new research from @OpenAI: How does CoT monitorability (no-CoT capability and CoT controllability) evolve over the model training lifecycle? We emerged with unsurprising but sobering findings: as we scale toward more intelligent systems, models are htt
VoxMel Wavelet-Factorized Autoregressive Mel-Spectrogram Modeling for Zero-Shot Speech Synthesis models: huggingface.co/VoxDance-Lab
🎊 Go 1.27.4 and 1.26.11 are released! 🔏 Security: Includes 12 security fixes to the standard library and the toolchain. 🔈 Announcement: groups.google.com/g/golang-announce/c/iY… ⬇️ Download: go.dev/dl #golang
Start building with Gemini 4 Carbon: our strongest model for complex, long-horizon tasks. It’s designed to explore problems thoroughly and run extended verification loops across software engineering, cybersecurity, knowledge work, and science.
With the rise of AI, this skill is more important than ever. I highly recommend this new book from @PrincetonUPress. As I wrote on the back, "Sharon Marcus knows what it takes to give a good talk, and she comes straight at you with it. She is as clear as she wants you to be."
MiniMax-H3-Image in 2 steps, realtime pDMD LoRA on MiniMax's 33B image model realtime editing on ZeroGPU LoRA + one-file ComfyUI checkpoint: huggingface.co/akhaliq/minimax-h3-image-… app: huggingface.co/spaces/akhaliq/minimax-h3…
I am pretty disappointed in Jensen as a person but this is indisputable. NVIDIA has been a champion of open research and open models, yes because it financially aligns with their interests (selling GPUs), but having a prominent champion in Jensen has been fantastic
Jensen no longer being CEO would be very very bad for the open model cause in the US.
Currently working towards a bigger open LLM roundup. But in the meantime, just one of the interesting trends that stood out. I.e., looks like the transition from full to hybrid attention continues. In 2025, there were still lots of full attention models, with a shift to "some
Five new VLAs in LeRobot 0.7. The zoo keeps growing 🤖 - G0.5 from @GalaxeaDynamics: a 2B VLA that writes its subtask, then its actions, in one stream. Works with /subtask and /autosteer - LingBot-VLA 2.0 from @robbyant_brain: one model across embodiments (55-D action space,
What does real-time transcription look like in a live setting? 🎙 We used Voxtral Transcribe 3 at AI Engineer Paris to capture every main-stage talk in real time. We boosted accuracy with context biasing, feeding the model speaker names and technical terms from the conference
Fast kernels, without the tuning wait: Helion's new compile-time heuristics automatically choose high-performance configurations, with no autotuning required. On H100 benchmarks, they beat handwritten Triton kernels for linear-attention forward passes by 18% and vLLM's
World In-Context Learning for Agentic Robot Manipulation paper: huggingface.co/papers/2611.12885
Imitation learning plateaus with more data. Additional benefit of collecting more data stops improving performance -- hence, Kay was convinced they needed reinforcement learning.
Done arguing over which type of molecule got life started, a new wave of researchers are deconstructing what life is, does, and could be — and are even trying to create it themselves. Read @yasemin_sap's feature on a new era for the research into the origins of life:
We launched PennyLane in 2018 to empower scientists and developers everywhere to build meaningful quantum algorithms. Today, it is one of the most utilized quantum software platforms. Thanks to everyone who has joined us on the journey so far. As we celebrate PennyLane’s 8th
"Its hard to tell the difference between a 92% model and a 99% model" -- so the bottleneck is evals.
vLLM relies heavily on PyTorch, with frequent calls to LibTorch in its C++ backend. As a result, changes to LibTorch’s binary interface may require vLLM to be rebuilt and relinked, even when the APIs remain unchanged. What if we could update LibTorch without rebuilding vLLM?
vLLM × PyTorch: with the LibTorch Stable ABI, new PyTorch releases no longer force a vLLM rebuild 📷 Users upgrade sooner, distributors rebuild less, devs test torch changes without recompiling vLLM. Read the migration story: https://t.co/bRhb5zK8Ez
Congrats to @MiniMax_AI on H3-Image taking the #1 spot among open-weight models for image editing by Artificial Analysis! 🎉 We’re bringing NVIDIA Vera Rubin support H3-image to vLLM-Omni, and native 4K generation in 0.87 seconds!⚡️ Check it out!👇 vllm.ai/blog/2026-11-12-minimax-h3-image…
ChatGPT required RL, it was not pure imitation learning or dagger; reasoning models also required RL. Obviously the same will be true of robotics, the question is just if base models are good enough
pendulum has fully swung back from where it was last year (“scale up data collection, behavior clone large models”). everyone is finally seeing the shining light of RL! https://t.co/ygTwbIvX4G
Hillclimbing: Make the model better (best intentions). Benchmarking: Measure how much better (best effort). Benchmaxxing: Make the numbers look better (best optics).
We gotta have a conversation about hill climbing vs benchmarking vs benchmaxxing
So many great world models out there for you to try
Today, we’re making Praxis-1 available to everyone. Our state-of-the-art world action model turns video pretraining into control for real robots, and it ships with open weights under Apache 2.0, ready to run and fine-tune on your own hardware. https://t.co/jL5ZmpY6x9
Ernie 5.5 flash preview huggingface.co/spaces/akhaliq/ernie-5.5-…
vLLM EPD, now upgraded 🚀 DeepInfer demonstrated the value of EPD disaggregation. vLLM’s latest EPD improvements deliver lower TTFT and higher throughput per GPU, counting all encoder GPUs, integrated with vLLM Router, SMG, and llm-d! Read more here!👇 vllm.ai/blog/2026-11-11-encoder-disaggre…
Gemini 4 Carbon is good at pelican SVGs, but that's nothing new for Gemini "Make a three js animation where yonkers stands up as a big mech (cool stand up motion) and stomps around manhattan and queens and Brooklyn, use openstreetmap mapping data" simonwillison.net/2026/Nov/10/llm-gemini
Lila Shroff @LilaShroff confesses she’s “recently become a bandwagon math fan.” She even manages to include a few correctly typeset polynomials in @TheAtlantic today, as she surveys the craziness in math and AI in the Bay Area and Silicon Valley. theatlantic.com/technology/2026/11/drunk…
A team of researchers mimicked hydrothermal vents in their lab to model the chemical reactions that may have helped life first form. quantamagazine.org/what-is-the-metabolis…
Praxis-1: an open-weight world action model for real robots 🤖 New from @runwayml and now in LeRobot. Praxis-1 learns how objects behave and how tasks unfold from large-scale video, which gives it a head start on a new robot. One checkpoint adapts with light fine-tuning or
This week at IBM #QDC26, we're introducing new tools that help developers build and scale quantum workloads, including new managed workflows and beta access to IBM Quantum Functions, bringing advanced ecosystem software directly to @IBM Quantum Platform. ibm.com/quantum/blog/dev-tools-qdc-2026
When your model is robust to Chinese adversaries, but you're not robust to adversarial Chinese models
New research from Microsoft Research demonstrates that reprogramming just a few dormant neurons in an AI agent can create a backdoor that leaves the agent behaving normally until it hits a specific trigger. While the specifics of this threat are novel, the mitigation is not: