MarkTechPost
AI news source. Articles are auto-selected and adapted by Hamidun News editors.
Latest publications
A Toolkit for the Claude API: The SuperClaude Framework
SuperClaude Framework gives developers built-in components for advanced AI workflows: commands, agents, operating modes and session memory, all in one system.
Memory for AI agents: Tencent's local system
Tencent has open-sourced TencentDB Agent Memory, a local memory system for AI agents that cuts token use by 61% and raises accuracy by 28%.
How NVIDIA's New Neural Network Saves Memory: Gated DeltaNet-2
NVIDIA has built Gated DeltaNet-2, a linear attention mechanism that manages memory in large language models better by using separate erase and write gates instead of a single gate.
Perplexity open-sources Bumblebee, an AI scanner to protect developer code
Perplexity has published the source code of Bumblebee, a tool that scans developer systems for vulnerable dependencies without running any code.
Gemini 3.5 for coding: a new model that is cheaper and faster
At I/O 2026, Google unveiled Gemini 3.5 Flash, a model that is 75% cheaper than the flagship, runs 4 times faster and handles code and automation better.
Alibaba releases a neural network for simultaneous video translation
Alibaba has unveiled a model that translates video and speech in real time across 60 languages, with minimal delay and while preserving the speaker's voice.
NVIDIA Nemotron text model offers three generation modes
NVIDIA has released Nemotron-Labs-Diffusion, a language model that combines three decoding modes and processes tokens 6 times faster than Qwen3-8B.
Knowledge Graphs for Text Analysis: How AI Extracts Information
A tutorial on automatically extracting entities and relations from text with kg-gen, building interactive knowledge graphs and analyzing them in NetworkX.
Turbovec: A Rust Vector Index Built on Google Research's TurboQuant Algorithm
Turbovec uses Google's TurboQuant algorithm to compress vectors 16x with no prior training, making RAG applications easier to deploy.
Best Agentic AI Platforms in 2026: Ranking Salesforce, Microsoft and Others
Companies are moving from pilots to production. MarkTechPost ranked the top 10 agentic AI platforms, including Salesforce Agentforce, Microsoft Copilot Studio and ServiceNow, with verified pricing and real deployment dat
NVIDIA develops a method for training neural networks in 4-bit precision
NVIDIA has unveiled NVFP4, a methodology for training large models in 4-bit precision instead of the standard 8-bit, halving memory use with no loss of quality.
OpenAI unveils MRC, a network protocol for supercomputers with millions of GPUs
OpenAI has created MRC, a new open networking protocol for large AI clusters. It spreads data across hundreds of paths and recovers from failures in microseconds, enabling supercomputers with 100,000+ GPUs.
Meta AI unveils NeuralBench, a framework for benchmarking brain-activity models
Meta has released NeuralBench, an open framework for standardized testing of EEG-based AI models, combining 36 tasks, 94 datasets and 13,603 hours of brain recordings in one interface.
How to Shrink a Language Model 3x: A Guide to FP8, GPTQ and SmoothQuant
A step-by-step guide shows how to compress large language models with llmcompressor, comparing FP8, GPTQ and SmoothQuant quantization to cut hardware demands.
OpenAI releases three audio models: translation, transcription and real-time reasoning
OpenAI has added three audio models to the Realtime API: a reasoning agent model, multilingual translation and streaming transcription.
Anthropic builds a tool that translates Claude's internal 'thoughts' into plain language
Anthropic has developed Natural Language Autoencoders, a technique that turns Claude's internal activations into text explanations, exposing how the neural network works.
NVIDIA Star Elastic: 360× Savings on Neural Network Training
NVIDIA has unveiled Star Elastic, a method that packs three models of different sizes into a single checkpoint and trains them 360 times more efficiently.
NVIDIA releases cuda-oxide, a Rust-to-GPU compiler
NVIDIA has open-sourced cuda-oxide, a tool that compiles Rust functions straight to PTX GPU code. It should simplify writing CUDA applications in Rust and make parallel computing more accessible.
NadirClaw: Cutting AI Costs Through Prompt Routing
NadirClaw is a prompt-routing tool that sorts requests into simple and complex ones and sends each to a suitable model to reduce costs.
Open-Source AI Agent Hermes Tops OpenRouter in Token Consumption
Nous Research's open-source AI agent Hermes has overtaken the closed OpenClaw platform to take first place on OpenRouter, generating 224 billion tokens a day. It took just three months.
Search Neural Networks in 2026: How to Choose Among Nine Systems
Nine leading vector databases are now core infrastructure for RAG and agentic AI. A review of architecture, price and scalability helps match one to your task.
AI for coding: Vercel's Zero language
Vercel has released Zero, a programming language whose compiler outputs JSON instead of text, so AI agents can write, fix and deploy native programs on their own.
How Memori Builds Persistent Memory for Agents and Multi-Session LLMs
Memori is a memory infrastructure for LLM applications that lets agents keep context across sessions and serve multiple users at the same time.
SHAP for Machine Learning: Comparing Explainers in a Practical Guide
A new guide shows how to choose among four SHAP methods for interpreting models, from the fast Tree explainer to the all-purpose Kernel.
DeepMind builds a Gemini-powered AI mouse so users don't have to switch to chat windows
Google DeepMind unveiled an AI mouse built on Gemini that reads the context around the cursor and takes voice commands without opening a separate window.
Mira Murati unveils first AI system for natural real-time conversation
Thinking Machines Lab has introduced TML-Interaction-Small, a model that listens and responds at the same time, processing speech, video and text without lag. It is the first AI that can converse like a human.
Nous Research unveils Lighthouse Attention to speed up LLM training
Nous Research has published Lighthouse Attention, a new attention mechanism that speeds up large language model training by 1.4–1.7x on long contexts.
Fastino Labs' GLiGuard: A Safety Model 16x Faster Than Much Larger Rivals
Fastino Labs has open-sourced GLiGuard, a 300-million-parameter moderation model that checks a prompt on four dimensions in a single pass and runs 16 times faster than comparable models.
Shadow AI at 63% of Companies: Tools Have Outrun Corporate Policy
Most companies still have no formal AI governance policies, yet employees are already adopting tools on their own, creating serious security risks.
Nous Research speeds up LLM pretraining 2.5x without changing the architecture
Nous Research has developed Token Superposition Training, a method that cuts language model pretraining time by 2.5x without changing the architecture or inference behavior.