Howdy! It’s Barsee again.

Happy Wednesday, AI family, and welcome back to AI Valley.

Today’s climb through the Valley reveals:

• Anthropic wins a major fair use victory for AI

• Google launches its open-source Gemini CLI

• Google’s DeepMind unveils AI robots that work offline

• Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

ABILITY AI

Stop talking about AI automation. Start building it

Many teams talk about AI, but they often struggle to implement it because of scattered tools, unclear workflows, and distant automation goals.

Welcome to the Ability AI Hub, a carefully curated workspace on Slack where business leaders, engineers, and automation experts can build and implement AI workflows. Powered by the Ability Hub Agent, you just describe a task – "automate onboarding," "summarize research," "analyze screenshots" – and get matched with real automation templates, setup guides, and live agents.

What's inside:

• 500+ tested automation templates

• Step-by-step setup guides in plain English

• Live support from the Ability AI team

The Hub Agent finds templates built for your specific tools and needs. For setup help or custom solutions, the Ability AI team offers professional services.

Join the Hub and create your first automation today!

_{*This is sponsored}

THROUGH THE VALLEY

1. Anthropic wins a major fair use victory for AI

A federal judge has ruled that Anthropic’s use of legally purchased physical books to train its AI models qualifies as fair use, a landmark decision favoring the AI industry. Judge William Alsup determined that digitizing print books for training was “spectacularly” transformative and likened it to teaching students to write better, not replicating or replacing authors. However, the court will hold a separate trial over Anthropic’s storage of millions of pirated books, which is not considered fair use. The ruling does not address AI output copyright issues, but sets a precedent for future training-data-related legal battles.

2. Google launches its open-source Gemini CLI

Google has launched Gemini CLI, an open-source AI tool built for terminal use, powered by the Gemini 2.5 Pro model. It allows developers to run coding, automation, research, and content generation tasks directly from the command line, using a massive 1 million token context window. It supports real-time web-grounded responses via Google Search and works in both interactive and scriptable modes. With a generous free tier (60 requests/min, 1,000/day), it's designed for individual devs and small teams. Gemini CLI competes directly with tools like Claude Code and OpenAI Codex, offering powerful, accessible AI tightly integrated into Google’s ecosystem.

3. An AI-powered robot can play badminton

Researchers at ETH Zurich have built a legged robot that can play badminton against humans using reinforcement learning. Named ANYmal-D, the robot uses a stereo camera and a dynamic arm to track, predict, and return shuttlecock shots in real time. Its unified controller combines vision, locomotion, and manipulation, allowing coordinated full-body motion. In tests, it achieved up to 10-shot rallies, adjusting its gait and posture to maintain balance. While it struggles with fast smashes due to hardware limits, the AI system itself is adaptable and has already been applied to robotic throwing. Future upgrades aim to reduce response latency for better gameplay.

4. SoftBank pitches $1 trillion US AI hub

SoftBank founder Masayoshi Son is reportedly planning a $1 trillion tech manufacturing hub in Arizona, dubbed “Project Crystal Land,” aiming to rival China’s Shenzhen. The industrial complex would focus on AI and robotics, and Son is seeking partnerships with giants like TSMC and Samsung. Talks are underway with U.S. federal and state officials for tax breaks, but the project’s future depends on political backing, particularly from the Trump administration. This follows SoftBank’s recent $6.5B acquisition of Ampere and a $40B commitment to OpenAI. If realized, it would surpass the $500B Stargate data center project in scale and ambition.

5. Thinking Machines Lab raises $2B at $10B valuation

Mira Murati’s new startup, Thinking Machines Lab, has raised a staggering $2 billion in seed funding at a $10 billion valuation, just six months after launching. The former OpenAI CTO, known for leading ChatGPT and DALL·E, has kept the company’s mission under wraps, but her reputation and a team of top AI researchers have attracted major investors. Andreessen Horowitz led the round, with backing from Conviction Partners. Murati left OpenAI in 2023 after clashing with CEO Sam Altman and briefly served as interim CEO during his short-lived ousting. Co-founder John Schulman and other ex-OpenAI staff have joined her team.

6. China’s Baidu to Open-Source Next-Gen Ernie model

Baidu is making a big pivot in its AI strategy by open-sourcing its next-gen Ernie model starting June 30. Once a firm believer in closed-source development, CEO Robin Li has shifted gears, likely influenced by the success of DeepSeek, an open-source Chinese startup competing with OpenAI-level performance. Baidu is also making its Ernie Bot free from April 1 to boost adoption. Despite being an early AI investor, Baidu trails rivals like ByteDance and DeepSeek in user numbers. Li now believes open access could accelerate AI adoption, with Ernie 4.5 launching soon and Ernie 5 planned for late 2025.

PEAK OF THE DAY

Google’s DeepMind unveils AI robots that work offline 🤖

Google DeepMind is ushering in a new era for robotics with the launch of an on-device version of its Gemini Robotics AI model. Unlike previous iterations, this compact and highly efficient version can run directly on a robot’s hardware, eliminating the need for a constant internet connection.

Here’s what you need to know:

• The Gemini Robotics model empowers robots to tackle a wide range of tasks, even those they haven’t encountered before. It enables machines to adapt to new environments, interpret complex commands, and perform intricate actions requiring fine motor skills.

• While the original Gemini Robotics model could operate both in the cloud and on devices, the new on-device version delivers nearly identical capabilities without relying on internet connections. This is a game-changer for settings with unreliable or restricted connectivity.

• According to Carolina Parada, Head of Robotics at Google DeepMind, the on-device model can master new tasks after seeing just 50 to 100 demonstrations. Initially trained on the ALOHA robot, it has already been adapted for use with other advanced robots, such as the humanoid Apollo and the dual-arm Franka FR3.

• For the first time, Google is releasing a software development kit (SDK) alongside the model, giving developers the tools to test, customize, and enhance the AI for their own robots.

• To ensure safety and reliability, the on-device model and SDK will initially be available to a select group of testers as Google continues to refine the technology.

Why it matters:

The on-device Gemini Robotics model represents a significant leap forward for autonomous robots, enabling them to process and act on information instantly, without waiting for cloud responses. This is especially valuable in locations with poor connectivity or strict security requirements. Its rapid learning capabilities and developer-friendly SDK open the door for innovation in industries ranging from manufacturing to healthcare. As Google DeepMind works to further strengthen safety, this model is poised to play a pivotal role in shaping the future of robotics, both online and offline.

TRENDING TOOLS

• Lambda 1-Click Cluster Technical Validation: Pre-configured, pre-validated AI clusters that deploy faster than your last Git push. *

• Airtable AI: The AI-native app platform, combining the magic of vibe coding business apps with real production-readiness and scalability.

• GPTHuman AI: Makes AI-generated text sound more human, helping users pass AI detection tests and avoid robotic tone.

• Pulse: Turn messy data into clear dashboards, just by chatting with AI.

• Shadow: Meet, decide, and get work done without looking back.

_{*asteric is sponsored}

THINK PIECES / BRAIN BOOST

• Why next-token prediction is enough for AGI.

• AI prompt engineering in 2025: What works and what doesn’t.

• The 20+ most common AI terms explained, simply.

THE VALLEY GEMS

What’s trending on social today:

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.