This should be interesting maven.com/p/1bb91a/reasoning…

🚨 AI BUILDERS — waitlist is OPEN 🚨 LLMs aren’t enough. Prompting isn't enough, Context engineering is NOT enough. "Thinking mode" is marketing bullshit. So I built ReasoningLayer a deterministic reasoning layer. And today… the waitlist is OPEN. 👉 reasoninglayer.ai/

General VLMs struggle with precise text localization & often hallucinate on dense docs. 🔥PP-OCRv5 solves this with a modular, two-stage pipeline: ✅Smaller, faster, lighter ✅Accurate bounding boxes ✅Edge-device friendly Read more 👇huggingface.co/blog/baidu/pp…

💥Big news! ERNIE-4.5-21B-A3B-Thinking just hit #1 overall on @huggingface trending models.🚀 👉Check out the model page: huggingface.co/baidu/ERNIE-4…

Also, while we have your attention... Flashcards & Quizzes are rolling out TODAY! 🥳 You can now create customizable flashcards and quizzes in @NotebookLM. Stumped on a question? Tap the Explain button to receive an in-depth summary in the chat. Study on!

Introducing Voxtral WebGPU: State-of-the-art audio transcription directly in your browser! 🤯 🗣️ Transcribe videos, meeting notes, songs and more 🔐 Runs on-device, meaning no data is sent to a server 🌎 Multilingual (8 languages) 🤗 Completely free (forever) & open source

Less than two weeks Kimi K2's release, @Alibaba_Qwen's new Qwen3-Coder surpasses it with half the size and double the context window. Despite a significant initial lead, open source models are catching up to closed source and seem to be reaching escape velocity.

Interesting new benchmark from Hugging Face testing how well vision LLMs can handle long video inputs (generally after they've been split into many thousands of images) - my notes here: simonwillison.net/2025/Jul/2…

465 people. 122 languages. 58,185 annotations! FineWeb-C v1 is complete! Communities worldwide have built their own educational quality datasets, proving that we don't need to wait for big tech to support languages. Huge thanks to all who contributed! huggingface.co/blog/davanstr…

Apple released DiffuCoder-7B a Masked Diffusion Model for Code Generation. MLX support is a work in progress right now.

RAG POCs are easy, but building production-grade retrieval is legitimately hard. These are things you don’t realize when you’re first starting out building agents - “wow my chat over 10 pdfs works in 10 mins!”. We learned these lessons as we built out LlamaCloud and wanted to share them with you 💡: 1. Access Control: if you want to gate data through user permissions, you need to invest in data connectors that actually index different levels of permission metadata within the source document. 2. Failure Planning: LLM APIs have rate-limits on large volumes. Data sources require you to think about credential expiration and data throughput. If you’re indexing 1M PDFs and it fails halfway through, ideally you don’t start from scratch. 3. Noisy Neighbors: If you have a lot of RAG pipelines through a central system, you need to make sure that one large job doesn’t take down every other pipeline. These days context engineering is a hot topic, and one of the primary sources of context remains your own enterprise data. Full credits to @thesourabhd for this thoughtful blog post: llamaindex.ai/blog/4-ways-ll… Come check out LlamaCloud if you want to have a managed document indexing/retrieval platform: cloud.llamaindex.ai/

Hand written RISC-V assembly 23.49x faster than C, thanks to work from @CAS__Science

Moonshot AI has quietly released their new coding model: Kimi-Dev 💻 @Kimi_Moonshot huggingface.co/moonshotai/Ki… ✨ 72B - MIT License ✨ 60.4% on SWE-bench Verified ✨ RL-trained to patch real repos in Docker ✨ Only rewarded if full test suite passes

December will mark 25 years of FFmpeg. Going from 100 downloads to being used by the NASA Mars Rover? What is in FFmpeg's future? What software do you still use that is 25 years old? Date: 2000-12-19 The first tarball is available! web.archive.org/web/20010710… link.springer.com/article/10…

🚨Breaking: New DeepSeek-r1 (0528) just tied for #1 in WebDev Arena, matching Claude Opus 4! More highlights: 💠 #6 Overall on Text Arena 💠 #2 in Coding, #4 in Hard Prompts, #5 in Math category 💠 MIT-licensed, currently the best open model on the leaderboard! Huge congrats @deepseek_ai on the incredible milestone 🔥

Everything will be local Yesterday I gave a talk about Real-World Applications of MLX and built a fast on-device semantic search index over Apple WWDC 2025 docs. Open-sourced the code for anyone curious! github.com/mcantillon21/loca…

✨Introducing ComfyUI-R1: a large reasoning model for automated workflow generation🌺 huggingface.co/papers/2506.0…

Introducing Qwen3! We release and open-weight Qwen3, our latest large language models, including 2 MoE models and 6 dense models, ranging from 0.6B to 235B. Our flagship model, Qwen3-235B-A22B, achieves competitive results in benchmark evaluations of coding, math, general capabilities, etc., when compared to other top-tier models such as DeepSeek-R1, o1, o3-mini, Grok-3, and Gemini-2.5-Pro. Additionally, the small MoE model, Qwen3-30B-A3B, outcompetes QwQ-32B with 10 times of activated parameters, and even a tiny model like Qwen3-4B can rival the performance of Qwen2.5-72B-Instruct. For more information, feel free to try them out in Qwen Chat Web (chat.qwen.ai) and APP and visit our GitHub, HF, ModelScope, etc. Blog: qwenlm.github.io/blog/qwen3/ GitHub: github.com/QwenLM/Qwen3 Hugging Face: huggingface.co/collections/Q… ModelScope: modelscope.cn/collections/Qw… The post-trained models, such as Qwen3-30B-A3B, along with their pre-trained counterparts (e.g., Qwen3-30B-A3B-Base), are now available on platforms like Hugging Face, ModelScope, and Kaggle. For deployment, we recommend using frameworks like SGLang and vLLM. For local usage, tools such as Ollama, LMStudio, MLX, llama.cpp, and KTransformers are highly recommended. These options ensure that users can easily integrate Qwen3 into their workflows, whether in research, development, or production environments. Hope you enjoy our new models!

LiveCC just dropped on Hugging Face Learning Video LLM with Streaming Speech Transcription at Scale video LLM capable of real-time commentary, trained with a novel video-ASR streaming method, SOTA on both streaming and offline benchmarks.

It's been 2.5 years with little progress finding mitigations for prompt injection attacks LLM apps... but that may finally have changed! Google DeepMind published a paper describing CaMeL, an ingenious system that could, maybe, lead to secure digital assistants