Show HN: DeepSeek V3 in ~200 lines of pure Jax
The architecture from the [DeepSeek-V3 technical report](https://arxiv.org/abs/2412.19437) in **190 lines of pure JAX**, pretrained from scratch on a MacBook Air on public-domain
The latest in AI, clustered and ranked. Repeated hype gets pushed down so the actual signal stays up top.
The architecture from the [DeepSeek-V3 technical report](https://arxiv.org/abs/2412.19437) in **190 lines of pure JAX**, pretrained from scratch on a MacBook Air on public-domain
* [English edition](https://www.infoq.com/news/2026/11/google-ambient-agent-prod/#) * [Chinese edition](https://www.infoq.cn/) * [Japanese edition](https://www.infoq.com/jp/) *
The code is at [https://github.com/higurashi-ai/Haru](https://github.com/higurashi-ai/Haru): a PyTorch implementation with a web demo, an MLX implementation with a native macOS
View a PDF of the paper titled The Little Book of Generative AI Foundations: An Intuitive Mathematical Primer, by Tianhua Chen
Many scientific simulations, from molecular chemistry to materials science, spend the majority of their execution time computing high-precision dense general matrix
A minimal JAX implementation of the DeepSeek V3 architecture in approximately 200 lines of code.
Guide to using NVIDIA NeMo Switchyard for cost-optimized model routing on Amazon Bedrock.
A browser-based interface designed for interacting with AI agents.
Case study on integrating no-code SageMaker Canvas models into production MLOps pipelines.
Anthropic CEO comments on compute costs and business uncertainty ahead of IPO.
Gemini 4 Carbon and Fable 5.2 outperform GPT-6.1 Sol on the MathArena benchmark.
Anthropic releases prompting guide for Claude Opus 5.5, recommending reduced effort settings.
OpenAI dissolves robotics team, shifting focus to general-purpose model control.
Runway releases Praxis-1, an open-source robotics model trained on video data.
An intuitive mathematical primer covering the foundational concepts of generative AI.
A technical exploration of transformer architecture through the lens of associative memory.
Amazon reportedly developing a comprehensive product knowledge graph for search and AI applications.
Analysis of open-source AI development independent of major hardware vendors.
Discussion on the potential impact of AI tools on ransomware attack sophistication.
Asian governments are assessing the impact of Jensen Huang's departure on chip supply and local AI projects.
Claims regarding Anthropic's improved training efficiency for the Mythos model.
Widespread flight cancellations across America may be tied to a new AI-powered air traffic management tool launched just hours ago. This video analyzes the link between recent radar issues and the dep
Guide to running native DLA inference on NVIDIA Jetson AGX Orin using the cuDLA Python API.
Engineering breakdown of using EC2 Spot instances and SQS to reduce inference costs for speech-to-text.
Technical guide on using vLLM to serve large models across multiple NVIDIA DGX stations.
Walkthrough on using SageMaker HyperPod Data Capture to log inference traffic for model fine-tuning.
Google open-sources AQuA, a tool for diagnosing production AI agent failures via session replay and root cause analysis.
Haru 4B, a small model designed for full-duplex speech-to-speech interaction.
OpenAI introduces ChatGPT Wallet, enabling AI agents to perform automated shopping and checkout for U.S. Pro users.
Deployment guide for OpenFold3 on SageMaker JumpStart for biomolecular structure prediction.
Implementation guide for attributing Amazon Bedrock inference costs to specific IAM principals.
Cloudflare adds local development support for Browser Run Quick Actions via wrangler.
Qwen 3.8 27B parameter conversational model released under Apache 2.0 license.
MiniMax H3 image generation model supporting text-to-image and image-to-image tasks.
LTX-2.5 diffusion model for video generation tasks including image-to-video and audio-to-video.
IDOL model for 3D reconstruction using Gaussian Splatting from images.
Moonshot AI Kimi K3.1 Pro conversational model with compressed tensor support.
Flux 3 image generation model optimized for ComfyUI workflows.
Meta releases SAM 3, a new iteration of the Segment Anything Model with video support.
ComfyUI integration for the MiniMax H3 image generation model.
ZONOS3-Mini text-to-speech model available in safetensors and GGUF formats.
Qwen 3 27B draft model specifically optimized for speculative decoding in SGLang.
NVIDIA updates cuBLAS with optimized FP64 algorithms to accelerate high-precision scientific simulations.
OpenAI launches ChatGPT Wallet for tokenized AI-powered shopping.
Technical patterns for reducing session startup latency in Amazon Bedrock AgentCore runtime.
LangChain reaches a $3 billion valuation, reporting 7,000 customers for its LangSmith platform.
Discussion on the limitations of self-verification in autonomous coding agents.
OpenAI introduces a wallet feature for its Dots AI agents to facilitate automated purchases.
Research indicates that advanced Chinese AI models were trained using export-restricted Nvidia chips.
ElevenLabs launches ElevenAI, a new application for enterprise meeting transcription and dictation.
Anthropic IPO excludes Chinese and Hong Kong investors due to US export controls.
This video installs and tests ZONOS3-Mini, Zyphra’s compact, open-source text-to-speech model. 📬Weekly AI Newsletter: https://fahdmirza.substack.com/ 🔥 Get 50% Discount on any A6000 or A5000 GPU ren
Community fine-tune of Qwen 3.5 9B model.
Nous Research fine-tune of Qwen 3.5 27B focused on value alignment.
OpenAI introduces a payment feature for agents in ChatGPT Work.
Google simplifies billing and credit management for Gemini API users in AI Studio.
Netflix is building an ontology-driven knowledge graph to improve observability for AI agents.
Inkling-2-Small MoE conversational model.
GGUF quantization of a fine-tuned Gemma 4 31B model.
Quantized version of Qwen-Image 2.1 for text-to-image tasks.
Mirage model for zero-shot object detection and image segmentation.
4-bit GGUF quantization of the Edge0 145B MoE model.
Jensen Huang has resigned from NVIDIA, and the bigger consequence is NVIDIA's next deal. Here's what changes for AI chip competition, AI infrastructure financing, export rules, and what you pay for AI
Cohere announces the acquisition of AISLE and integration with cubic for enterprise AI services.
FAA expands use of AI tools for managing air traffic congestion across US airspace.
Same signal-first ranking, earlier dates.
Open archive