
Google releases open‑source Gemma 4 LLM under Apache 2.0 license
Google announced Gemma 4, an open‑source large language model available under the permissive Apache 2.0 license. The suite offers four variants, from a 2‑billion‑parameter edge model to a 31‑billion‑parameter dense version, and supports multimodal inputs with a context window up to 256 K tokens.
Also developing:
By the numbers: Antaris secures $28M Series A to accelerate AI‑driven analytics

The video “The dark truth behind AI – ‘Anatomy of an AI System’” argues that artificial‑intelligence systems are fundamentally opaque, both technically and legally, and that understanding them requires digging through multiple hidden layers. It points out three “black boxes”: the algorithmic opacity where even developers cannot fully explain model behavior; the corporate secrecy that shields data sources, training methods, and intellectual property; and the geographic supply‑chain black box that obscures where raw materials and labor originate. The presenter uses the Amazon Echo as a case study, quantifying the minerals, energy, water and human labor needed to manufacture a single device and to train generative models. “It looks like a technical drawing but, in a way, it’s a highly political, highly philosophical construction of extraction, of exploitation of inequality, of exploitation of human labour, of extraction of our private data,” the narrator says, highlighting how AI’s material footprint intertwines with social inequities and data exploitation on a planetary scale. The implication is clear: without greater transparency and accountability, AI’s growth will deepen environmental strain and widen socioeconomic gaps, prompting regulators, investors, and consumers to demand sustainable, ethical supply‑chain disclosures and robust oversight mechanisms.
Wild timing on this drop. We just spent weeks reverse-engineering how top agents actually store and retrieve memory. ByteRover's answer: file-based hierarchy, >92% retrieval accuracy, ~1.6s avg retrieval time, and 50-70% token savings even on lightweight models. Persistent memory that developers can actually...

Agnieszka Kurant frames art as a living organism powered by the collective intelligence of its audience. She describes a system where human‑generated digital traces—social media posts, protest chants, emotional reactions—feed directly into AI‑driven installations, turning viewers into co‑creators. The project scrapes...

Fabien Giraud’s project The Feral treats AI development as a child‑raising exercise, constructing a meticulously imagined world that recreates life on a French hill a millennium ago. By presenting this fictional environment to the machine, he aims to teach it...

AI is becoming your new customer. Visa reports a shift from B2H to B2AI, where AI agents compare, negotiate, and buy for people. 40% of Americans bought something through AI influence, and 53% of businesses will let AI negotiate deals. Yet 60% refuse...
Exploring the capabilities of AI live chat reveals diverse options for engaging conversations. With various platforms available, it raises the question: Is ChatGPT the top choice? https://t.co/bkPA4FnvFN

Meta’s announced $2 billion purchase of Chinese AI startup Manis has drawn intense scrutiny from Beijing. The China Commerce Ministry, through spokesperson He Hadong, reiterated that cross‑border business is permissible only when it complies with Chinese law, prompting a formal review...
Decision traces are a big deal and now possible with 1M token context *sneak peek: one of my projects is launching soon and will be focused on decision traces in agentic engineering
You can now choose between speed and quality for Imagine. Professional mode is coming later this month.

The video outlines a four‑tier framework for AI proficiency—unacceptable, capable, adaptive, and transformative—and argues that firms should prioritize these skill levels over traditional metrics like GPA when hiring new talent. Jensen Huang’s remark that graduates are judged on AI ability sets...

Microsoft unveiled the open‑source Agent Governance Toolkit, a seven‑package suite that adds policy, identity, compliance and reliability controls to autonomous AI agents. The toolkit runs in Python, TypeScript, Rust, Go and .NET, offering sub‑millisecond policy enforcement and cryptographic trust mechanisms....
AI sales agents are software that automate routine sales tasks such as abandoned‑cart recovery, lead qualification, and personalized follow‑ups, allowing merchants to operate 24/7 without expanding staff. They connect to ecommerce platforms, CRMs, and email tools via APIs, executing actions...

UBTech Robotics, a Chinese humanoid robot maker, announced a chief scientist position with a compensation package up to 124 million yuan (about $18 million) annually. The role will define the company’s humanoid robot and embodied intelligence roadmap and lead AI model research....
Google has launched its AI Plus subscription in 35 countries, including the U.S., bundling enhanced AI tools, cloud storage, and video generation features. https://t.co/Md1MWhHKRE
Now this is really interesting. It gets at whether AI models will be able to not just push at the edges of human knowledge but actually take a leap beyond the current human understanding into paradigm shifting breakthroughs. No clear...

Asia‑Pacific still bears roughly 63% of global workplace fatalities, with construction and manufacturing driving the high death rate. Regulators across the region are turning safety tech into law, exemplified by Singapore’s mandate for video‑surveillance on projects exceeding US$3.9 million and heightened...
Cyber attacks launched by malicious humans using AI are a very real AI risk. The best way to guard against them is to use AI to proactively find vulnerabilities in our systems and harden them, along with parallel efforts in...

Microsoft announced a four‑year, $10 billion investment program to accelerate artificial‑intelligence adoption in Japan. The plan will build cloud and AI infrastructure in partnership with Sakura Internet and SoftBank, which will provide GPUs and other compute resources. Sakura Internet’s stock surged...
Anthropic has imposed stricter five‑hour usage limits on its Claude chatbot and Claude Code coding assistant, prompting angry complaints from developers. The move comes as the company grapples with a recent accidental source‑code leak that sparked a wave of community‑driven...
Which company’s dollars are more likely to lead an advance in AI for science? 🤔