MarkTechPost
AI news source. Articles are auto-selected and adapted by Hamidun News editors.
Latest publications
Zyphra releases first MoE diffusion model, with a 7.7x inference speedup
Zyphra unveiled ZAYA1-8B-Diffusion-Preview, the first MoE diffusion model converted from an LLM. Inference runs 7.7x faster by shifting work from memory to GPU compute.
How to Build an MCP-Style System for Intelligent AI Agents
A complete walkthrough for building an AI agent system that discovers the tools it needs, routes tasks and executes them step by step.
BerriAI releases LiteLLM Agent Platform for running agents in production
BerriAI has open-sourced LiteLLM Agent Platform, a Kubernetes-based infrastructure for running AI agents in production with isolated environments and sessions that persist across restarts.
Repowise for repository analysis: graph analysis and dead code detection
Repowise helps developers analyze code repositories at the whole-project level: dependency graphs, dead code detection and context for AI.
NVIDIA unveils SANA-WM: a model for 60-second 720p video on a single GPU
NVIDIA has shown SANA-WM, an open world model that generates 60-second 720p videos with precise camera control on a single graphics processor.
How to Build a Streaming Decision Agent with Online Replanning in a Dynamic Environment
A detailed tutorial on building a streaming decision-making agent that adapts to moving obstacles and shifting goals in real time.
Structured AI Output: How to Use Outlines and Pydantic
A new guide shows how to build LLM pipelines with a strict schema, typed data, JSON repair and validated function calls using Outlines and Pydantic.
Volcengine releases OpenViking, a file-based context database for AI agents like OpenClaw
Volcengine's OpenViking stores an AI agent's memory, resources and skills as a file system, aiming to simplify context retrieval, cut token use and make RAG navigation observable.
Neural Networks for Physics Simulation: A JAX and Neural ODE Guide
A new Diffrax and JAX walkthrough covers the full path: setting up the environment, adaptive ODE solvers, stochastic simulations, vmap batching and neural ODE training.
AI-Powered Search: How RAG Differs from Traditional Search
BM25 ranks documents by word frequency and rarity, while RAG with embeddings searches by meaning. That's why production systems increasingly combine the two.
Tribe v2: Meta's Neural Network for Analyzing Video, Audio and Text
Meta has open-sourced Tribe v2, a model that predicts the brain's fMRI response to video, audio and text, giving neuroscientists a tool for virtual experiments without new scans. *Meta is designated an extremist organiz
Google ADK Demo Shows a Multi-Agent Python Pipeline for Data Analysis, Charts and Reports
A new guide shows how to build a multi-agent system on Google ADK that loads tables, runs statistical tests, draws charts and generates reports automatically in Python.
Google lets Gemini API combine Search, Maps and custom functions in a single request
Google has updated the Gemini API: one call can now combine Google Search, Google Maps and custom functions, enabling multi-step agent workflows without manual glue code.
OpenClaw: Building a Local, Locked-Down Runtime for AI Agents Without the Cloud
A new OpenClaw guide shows how to set up a local-first runtime for an AI agent: a loopback-only gateway, model access via environment variables, controlled exec and deterministic skills.
Microsoft shows how to run VibeVoice for ASR, realtime TTS and speech-to-speech
Microsoft has released a hands-on Colab guide to VibeVoice covering speaker-aware ASR, realtime TTS, batch audio processing and a speech-to-speech pipeline.
Hugging Face and Lambda: How to Parse and Fine-Tune on Agent Reasoning Traces
MarkTechPost has published a practical guide to the lambda/hermes-agent-reasoning-traces dataset: parsing agent thoughts and tool calls, building charts, and preparing data for fine-tuning.
Alibaba releases Qwen3.6-35B-A3B, an open multimodal MoE model built for agentic coding
Qwen has released open weights for Qwen3.6-35B-A3B, a 35B-parameter MoE model with 3B active that handles code, images and video, with a focus on tool use and agents.
Microsoft, NVIDIA and IBM Make the List of 2026's 19 Top AI Red Teaming Tools
A new AI red teaming roundup covers 19 tools, from Microsoft PyRIT and NVIDIA Garak to IBM AIF360, for catching data leaks, jailbreaks and model bias before production.
Meta unveils Autodata, an agentic system for building high-quality training data
Meta has introduced Autodata, a framework in which LLM agents collect, verify and refine training data themselves, so that strong models learn from harder, more accurate examples. *Meta is designated an extremist organi
NeuralSet and Deep Learning: Decoding Language Features from Brain MEG Signals
A new walkthrough with NeuralSet shows how to turn raw MEG signals into a model that predicts language features, including word length, directly from brain activity.
Microsoft OpenMementos: Working With Context Compression and Training Data
A step-by-step walkthrough of Microsoft's OpenMementos dataset shows how to stream reasoning traces, parse blocks and mementos, measure context compression and prepare fine-tuning data.
Stanford unveils OpenJarvis, a stack for local AI agents with memory and learning
Stanford has unveiled OpenJarvis, a platform for personal AI agents that run on-device, use memory and tools, and learn from local data.
Garry Tan releases gstack, a workflow system for Claude Code with QA, review and release modes
Garry Tan has open-sourced gstack, a set of modes for Claude Code that separates planning, review, QA and release into distinct commands and adds a persistent browser.
Zhipu AI releases GLM-OCR, a compact 0.9B-parameter OCR model for documents
Zhipu AI and Tsinghua University unveiled GLM-OCR, a 0.9B-parameter multimodal OCR model that parses complex documents, tables and formulas without betting on sheer size.
LangChain releases Deep Agents for multistep AI agents with memory and isolation
LangChain has released Deep Agents, a library for AI agents that need planning, long-term memory, file context and subagent isolation in long-running tasks.
IBM releases Granite 4.0 1B Speech, a compact multilingual speech model for edge AI
IBM has introduced Granite 4.0 1B Speech, a lightweight model for speech recognition and translation built for fast deployment in resource-constrained enterprise and edge scenarios.
Moonshot AI unveils Attention Residuals, an alternative to residual connections in transformers
Moonshot AI proposes Attention Residuals, a replacement for standard residual connections in transformers that lets each layer pick useful signals through attention over network depth.
Mistral releases Small 4, a 119-billion-parameter MoE model for reasoning, code and multimodal work
Mistral Small 4 combines instruct, reasoning, coding and image understanding in one open-source MoE model with 119B parameters and a 256k context window.
Google releases WAXAL, an open speech dataset for African languages
Google has opened WAXAL, a large speech corpus for African languages meant to speed up speech recognition and synthesis where open data has been critically scarce.
Nvidia open-sources OpenShell, a secure runtime for autonomous AI agents
Nvidia has released OpenShell as open source: a runtime with sandboxes, access policies and private routing, built to let autonomous AI agents work with code and the network more safely.