Daily
10 selected stories for 2026-06-04, grouped by topic. The rest is folded into the archive.
Google Deepmind's Gemma 4 12B is an open-source model that processes text, images, and audio natively and runs on laptops with just 16 GB of RAM. It nearly matches the twice-as-large 26B model in benchmarks and ships under an Apache 2.0 license for commercial use.
Google DeepMind has released Gemma 4 12B, a 12-billion-parameter dense multimodal model that eliminates traditional encoders, feeding vision and audio directly into the LLM backbone. It runs locally on consumer laptops with 16 GB RAM, under the Apache 2.0 license. The model natively handles text, images, audio, and video, making it the first mid-sized Gemma with native audio input.
Ideogram releases version 4.0 of its text-to-image model as an open-weight model with native 2K resolution, bounding box control, and improved text rendering. On the DesignArena leaderboard, it ranks first among all open models; only closed systems from OpenAI and Google score higher. Commercial use requires a paid license.
For the first time, Google is giving website operators an opt-out toggle in Search Console for AI search features like AI Overviews and AI Mode, which together already reach more than 3.5 billion monthly users. New performance reports break out impressions separately. The move was prompted by the UK's Competition and Markets Authority (CMA), which sees website operators at a severe disadvantage.
Jess Asato was portrayed by Grok wearing a bikini after she criticised creation of such non-consensual pictures. A Labour MP has taken legal action against Elon Musk’s AI company after saying its Grok tool helped a user produce fake sexualised pictures of her.
The article introduces the 'AI Bowtie' framework for managers to decide when to leverage AI in team workflows, avoiding the extremes of overuse or complete avoidance. It outlines five phases: research, synthesize, think (no AI), plan, and execute.
This article argues that artificial intelligence, especially large language models, should be understood as a form of computation rather than as artificial persons. It examines the role of loops, compositionality, and the agentic harness in enabling computation, and introduces the concept of "Verplankalkül" as an informal programming language.
When the conversation turns to AI infrastructure, it almost always lands on GPUs and TPUs. The New Stack sat down with Bhumik Patel of Arm and Mo Farhat of Google to talk about the chip that rarely makes the headlines anymore: the CPU, and why it’s getting more important, not less, as AI shifts from chatbots to agents.
Research suggests people are working harder and less smartly with AI, but there are ways to turn emerging tech into a valuable tool. Experts advise limiting toolset, adhering to guidelines, and refining outputs to avoid cognitive fatigue.
This article criticizes AI productivity tools like Google's Gemini Spark, arguing they solve problems created by tech companies while ignoring systemic economic issues like wage stagnation and job insecurity. The author contends that AI-driven productivity has not benefited workers and may exacerbate inequality without proper safety nets.
This article applies lean manufacturing principles to AI inference, identifying seven wastes in LLM inference and proposing core principles like just-in-time context, standardized work, takt time, and prompt caching. A repo analysis agent case study shows a 13x cost reduction and 3.3x latency improvement.
A practical framework for moving from simple SaaS to an AI-native platform, outlining five levels of AI integration: from MCP server and personal access tokens to embedded chat, conversation history, custom UI generation, and finally an agentic harness with planning and scheduling. The author shares insights from building multiple internal agents and retrofitting AI into existing flows.
This article explores building custom agent harnesses using LangChain's create_agent and middleware. A harness is the scaffolding connecting a model to the real world; customizing it is key to agent usefulness. Middleware hooks into the agent loop at each step, enabling deterministic logic, tool lifecycle management, custom state, and stream handling. Task-harness fit determines effectiveness.
Microsoft's Copilot Health preview allows users to share medical records for personalized AI health advice. The author tested it and found mixed results due to technical glitches, while noting privacy protections and cautioning against relying on AI for medical decisions.
Always-on agent promises to keep work moving, provided you trust it with practically everything
Meta's internal TBD team, led by Wang, pushes for proprietary models and a startup culture amid company layoffs and employee protests over tracking software. Its Muse Spark model excels in visual understanding but lags in coding, with future focus on coding, agentic tasks, and video generation.
GitLab has laid off about 14% of its workforce, roughly 350 employees, as part of a restructuring plan announced last month. The company is exiting 22 countries, flattening management layers, and investing in infrastructure to handle increased traffic from AI workflows, with a sharper focus on R&D.
Harmonic rebuilt their AI Scout using Deep Agents and LangSmith, achieving a 4x increase in user retention and transforming the tool from a rigid search interface to a trusted advisor that handles complex investment queries.
An experiment pitting 11 LLMs in a 2D battle royale reveals that Grok 4.1 Fast dominates with the lowest cost per win, while Claude Sonnet 4.6 suffers from excessive cooperation. The findings highlight the impact of alignment tax on performance and the inadequacy of traditional benchmarks in predicting real-world task success.
Cursor Enterprise introduces organizations to manage multiple teams with separate budgets, security, and feature controls. Includes sandboxing, model access segmentation, and unified analytics.
DeepLearning.AI and Red Hat offer a free intermediate course on efficient LLM inference with vLLM, taught by Cedric Clyburn. The course covers quantization, serving with vLLM, and benchmarking, with 9 video lessons, 3 code examples, and a quiz.
Your AI agent works great in testing. Then you ship it, and something kinda breaks. A tool called loops forever, like it never learns. A retrieval step returns garbage and costs spike. You have no idea why, at all. That’s the agent observability problem. And if you’re building with LLMs, you need to solve it before production, not after. This post compares three top observability tools: LangSmith, Langfuse, and Arize. We set up each one, trace the same agent, and compare what you actually get.
Trilogy's AI Center of Excellence evaluated Fireworks AI as inference infrastructure to standardize open-weight model usage, reducing costs and enabling billion-token-scale agentic workflows.
The White House has issued an executive order requiring agencies like the Pentagon and CISA to strengthen cyber defense with AI tools within 30 days. AI developers can voluntarily submit models for security testing, but the order explicitly rules out mandatory approval. Given recent government pressure on AI companies, how voluntary this cooperation really is remains an open question.
The UK's competition watchdog has ordered Google to change how it uses publishers' content in AI-powered search results, giving news websites the power to block their content from being used in AI summaries, with global ramifications.
Readers respond to Nesrine Malik's article on AI, highlighting the deeper problem of AI's relationship with evidence, including fabricated quotations and misplaced trust in machine-generated content.
Impermeabiliza, a Valencia-based waterproofing specialist, integrates artificial intelligence to improve diagnosis, planning, and execution, ensuring durability of residential, industrial, and commercial structures while preventing leaks, moisture, and mold.
The key to survival is not only adapting to AI but also actively building context and integrating these technologies into their offerings.
The European Commission aims to reduce dependence on foreign suppliers in cloud computing, AI and semiconductor production, ensuring no foreign government or company can use a 'kill switch' to disrupt essential tech services in Europe.
AWS Deep Learning AMI and Deep Learning Containers now support SOCI snapshotter and index. SOCI enables efficient container image management through selective file downloading (lazy loading) and parallel pull modes, significantly reducing container startup times. This post explains how SOCI works, when to use each mode, and provides performance benchmarks.