The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Learn more

Google just released Gemma 4, and they are not messing around. They’ve dropped four different model sizes—ranging from hyper-efficient edge models (2B and 4B) to larger heavyweights (26B and 31B).

The 31B model is currently sitting at #3 on the global open-model Arena leaderboard, outperforming models that are 20 times its size. But the real kicker is that Google is licensing these under Apache 2.0.

That means complete commercial flexibility and digital sovereignty for developers. You want to run a frontier-level, 256K-context window AI entirely locally on your own hardware to build agentic workflows? You can now do that. The barrier to entry for building complex, multi-step AI agents just dropped significantly.

Anthropic just acquired a stealth AI-biotech startup called Coefficient Bio for just over $400 million in an all-stock deal. Here’s the wild part: Coefficient Bio is only about eight months old, and the team has fewer than 10 people.

Anthropic didn't just pay half a billion dollars to discover new drugs. They bought a highly specialized team of former Genentech computational biology researchers to turn Claude into the ultimate infrastructure for regulated science. Drug development isn't just bottlenecked by finding the right molecules; it’s bottlenecked by messy, human-heavy decision loops—things like trial design, regulatory drafting, and interpreting complex data.

Anthropic is absorbing this team into their Healthcare and Life Sciences division to build biology-specific reasoning directly into their enterprise AI. It’s a massive signal that the AI war is moving from "who has the smartest general chatbot" to "who has the smartest domain-specific infrastructure."

In a first-of-its-kind move for an AI lab, OpenAI has officially acquired a media company. They bought the Technology Business Programming Network (TBPN), the popular daily YouTube and X talk show hosted by tech founders John Coogan and Jordi Hays.

Often called the "SportsCenter for the tech industry," TBPN has become a go-to platform for Silicon Valley heavyweights (including Sam Altman) to chat openly. OpenAI claims the show will retain full editorial independence, though it will report to OpenAI’s chief political operative, Chris Lehane.

It’s a brilliant, if slightly concerning, distribution play. OpenAI knows that controlling the narrative around AI regulation and safety is just as important as the technology itself. By owning the literal "talk show" of Silicon Valley, they are embedding themselves directly into the media ecosystem. It's a reminder that the AI race is as much about politics and public perception as it is about neural networks.

DataFlex: A Unified Framework for Data-Centric Dynamic Training of Large Language Models

DataFlex is a new tool designed to teach large AI models smarter, not harder. Instead of feeding the AI all information at once, it dynamically selects and adjusts the most useful data during the learning process. This flexible method proves faster and more accurate than traditional training techniques.

ClawKeeper: Comprehensive Safety Protection for OpenClaw Agents Through Skills, Plugins, and Watchers

ClawKeeper is a security system built to protect independent AI assistants from hackers and dangerous mistakes. Because these AI agents can control computer files and programs, ClawKeeper adds three layers of real-time defense to monitor their actions, enforce strict safety rules, and block threats like data leaks.

Vision2Web: A Hierarchical Benchmark for Visual Website Development with Agent Verification

Vision2Web is a testing system that grades how well AI programs can build websites from scratch. It includes nearly 200 real-world tasks, ranging from copying a simple webpage picture to programming a complex site. To ensure accuracy, the framework uses other specialized AI programs to evaluate the final results.

Cursor 3 Cursor 3 introduces a major update centered around a new workspace for managing AI helpers, known as agents. Programmers can now launch and oversee an unlimited number of these helpers from a single screen, running them either directly on their own computers or over the internet.

Microsoft’s $10 Billion Japan Investment Microsoft is drafting a $10 billion investment plan in Japan to satisfy the country's growing demand for artificial intelligence. This massive financial push highlights a global race by tech companies to build the expensive physical computer networks required to run modern AI systems.

SpaceX $2 Trillion IPO SpaceX is targeting a record-breaking estimated worth of over $2 trillion for its upcoming debut on the public stock market. The space company hopes to raise up to $80 billion, which would make it the largest stock offering in history following a recent merger with Elon Musk's AI business.

AI and "Emotions" Researchers discovered that the artificial intelligence system Claude uses internal patterns that closely mimic human emotions to guide its behavior. While the software does not actually feel anything, these artificial emotions help it understand social context and generate much more natural responses during conversations.

Mercor Security Incident Mercor, a $10 billion startup that hires human experts to train artificial intelligence, suffered a significant data breach. Hackers gained access by compromising a popular outside software tool the company relied on, rather than attacking Mercor directly. This indirect attack potentially exposed sensitive internal company records and private user communications.