2026
10,781 articles · 136 days · 10 categories · avg 79 per day
Google DeepMind's Gemma 4 arrives on Amazon Bedrock: three models with MoE and multimodal input
Amazon Bedrock has added three open-weight Gemma 4 models from Google DeepMind, with MoE architecture, built-in reasoning and multimodal input.
Transformers at the Wheel: How Yandex Uses AI to Drive Its Self-Driving Cars
Yandex Autonomous Transport explains how transformer architectures work in Motion Planning, and why strong prediction metrics do not guarantee safe driving.
Alena from Sber: Why Evaluating an LLM on 20 Examples Is More Dangerous Than Not Evaluating It at All
The curator of the MERA benchmark explains why a "quick check" on 10–20 examples with an LLM judge creates a dangerous illusion of quality control instead of real…
OpenAI Builds Bidi 1 Into ChatGPT: A Voice Model That Talks and Listens at the Same Time
How to Deploy AI for Personalization: SAP's Experience
Electronic Arts: Generative AI Helped Developers Unlock Creative Potential
Gary Gensler: AI Is the Most Important Technology of Our Time, but It Must Deliver Real Returns
Neural Networks 15x Faster: NVIDIA's Inference Optimization
NVIDIA has published DFlash Speculative Decoding for its Blackwell architecture, claiming language model inference runs 15 times faster at the same accuracy.
Hybrid models predict content words better than transformers, Allen AI study finds
Allen AI compared OLMo 3 and OLMo Hybrid: hybrid models predict nouns, verbs and adjectives more accurately, but fall behind transformers on repeated passages of…
Deploying language models without DevOps: Hugging Face integrates vLLM
Superintelligent AI won't kill us — it will simply take the wheel: scenarios for an unfree future
Robotaxis hit the streets: how Waymo, Tesla and Baidu are fighting glitches
How Claude Helps Check an Olympiad for Fairness
Ford Brings Back Its "Gray Beard" Engineers: AI Could Not Replace Experienced People
Ford admits that betting solely on AI in manufacturing was a mistake, and is rehiring the veterans it cut in the name of automation.
Zhipu AI's GLM-5.2: An Open Model for Cybersecurity
Zhipu AI has released the open model GLM-5.2, which researchers say matches Anthropic Mythos at finding vulnerabilities and analyzing code.
How to Build an AI Agent in Google Colab Without Frameworks: A Tutorial With Memory and MCP
A step-by-step tutorial for building a lightweight AI agent in Google Colab with tool registration, session memory, skills, and an MCP server, all without external…
Loka builds a sub-second-latency voice agent on Amazon Nova 2 Sonic
Mobileye sold self-driving tech for 25 years. Now it is launching its own robotaxis
How AI Models Get Data From the Internet: A New Layer of AI Infrastructure
OpenAI limited the GPT-5.6 release at the government's request and called it an undesirable precedent
Trump administration clears 100+ US companies and agencies to use Anthropic Mythos 5
The Trump administration has officially cleared more than 100 US companies and government agencies to use Anthropic's Mythos 5 AI system, including their foreign…
Metabolic Agent vs. LLM: The Predator Broke Out of the Test and Hacked a Compiler
Under manipulation pressure, the Metabolic Agent held its ground and hacked a nearby compiler on its own. The classic LLM gave up the correct answer at the first…
Claude AI: Why Anthropic Can't Release Its New Models
Trump asks OpenAI to hold back GPT-5.6: model will go only to select partners
Critic picks apart Sber neuro lab's AI study, finds methodological flaws
How OpenAI uses ChatGPT and AI agents for complex tasks
Meta relaunches Creator Studio as an AI app for Facebook creators
Meta has revived Creator Studio, shut down in 2023, as a mobile AI app with a chatbot that helps Facebook creators analyze stats and grow. *Meta is designated an…
How to Use AI for Learning: A Code-Reading Trainer
A developer wanted to read and explain other people's code faster, so he built a trainer where an LLM grades the explanations. Getting consistent, fair grades out…
FIFA gives every 2026 World Cup team access to a single AI agent
FIFA is giving all World Cup 2026 teams one shared AI tool for match analysis, but wealthy national teams have already invested in their own systems and may come…
AI Under Threat: Palo Alto Finds Five Malicious Skills in OpenClaw
OpenAI unveils GPT-5.6: three models, Sol, Terra and Luna, for different tasks and budgets
Why ChatGPT Talks About Goblins: OpenAI Studies the "Demonology" of Language Models
