Good morning, AI enthusiasts. "A huge change is coming," and those who learn how to upskill themselves "continually" will be the ones thriving in this AI-driven future.

These words from Google’s Demis Hassabis set a new standard for skill development, but the question remains: what frameworks can workers and students use to realistically keep pace when the AI space is evolving almost every week?

In today’s AI rundown:

• Demis Hassabis: AI demands ‘continual’ learning

• China's 'brain-like' AI runs 100x faster on its own chips

• Create enterprise-ready sounds with Stable Audio 2.5

• Harvard’s AI helps reverse disease in cells

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

AI LEARNING

🤖 Demis Hassabis: AI demands ‘continual’ learning

Image source: Getty Images

The Rundown: The top skill for the next generation in an AI-driven world? Learning how to learn. Speaking in Athens, Demis Hassabis, CEO of Google DeepMind, said mastering this skill is crucial as AI reshapes education, work, and industries.

The details:

• Hassabis warned that the pace of AI change is so fast that “the only thing you can say for certain is that huge change is coming.”

• He added that AGI (when AI matches humans at most tasks) could be achieved in a decade, bringing dramatic advances and a future of “radical abundance.”

• Thriving in this era will require meta-skills—the ability to continually optimize the learning approach to new subjects—alongside traditional knowledge.

• Hassabis added that this phase of lifelong learning is unavoidable, noting he is sure that people will have to continually learn “throughout their careers.”

Why it matters: The idea of learning a single skill and working with it for 5–10 years before needing to upskill is changing. In an AI-driven world, workers and students will need to constantly track developments in their domain, find ways to stay ahead of the curve, and differentiate themselves to build successful careers.

TOGETHER WITH VANTA

🛡️ Building trust in the AI boom

The Rundown: In today's AI boom, shipping fast gets attention — but building trust gets results. Vanta and Mercury show how SOC 2, audit-ready financials, and risk controls have become day-one signals that fuel growth, not later-stage checkboxes.

In this virtual event, learn to:

• Pass investor diligence with confidence and fewer follow-ups

• Avoid security and procurement blockers before they stall deals

• Stay ahead of compliance without hiring full teams

Register now to build enterprise-grade credibility without slowing down.

CHINA

🧠 China's 'brain-like' AI runs 100x faster on its own chips

Image source: Ideogram / The Rundown

The Rundown: Chinese researchers just published new research on SpikingBrain 1.0, an AI system that mimics human neurons to achieve massive speed gains while running entirely on China’s MetaX chips instead of Nvidia hardware.

The details:

• The model fires neurons selectively, similar to human brain activity, rather than activating entire networks like ChatGPT does continuously.

• Researchers trained 7B and 76B versions using under 2% of the data required by traditional models, matching their performance on language tasks.

• Tests showed the smaller model processed a 4M-token prompt over 100x faster than standard systems while maintaining stability for weeks.

• The Beijing team released a free version online where users can test "Shunxi," emphasizing it runs entirely on Chinese tech without any Western components.

Why it matters: By achieving massive speed gains on fully domestic hardware, China shows it can not only compete with but also potentially bypass largely Nvidia-powered Western AI. The landscape is no longer dominated by one ecosystem, making awareness of regional innovations essential for staying ahead.

AI TRAINING

🎶 Create enterprise-ready sounds with Stable Audio 2.5

The Rundown: In this tutorial, you will learn how to generate professional music and soundscapes for marketing, ads, and branded content using Stable Audio 2.5 on Replicate, creating enterprise-ready tracks at just $0.20 each.

Step-by-step:

1. Go to Replicate.com and search for "Stable Audio 2.5" by Stability AI (fund your account with a few dollars to run the model)

2. Set your parameters: choose genre (pop, hip-hop, orchestral), specify tempo in BPM, set duration up to 90 seconds, and adjust steps for quality

3. Write detailed prompts like "atmospheric pop electronic fusion with crisp percussion and layered synths" for cleaner results

4. Click Run to generate your track in under 10 seconds, then download for unlimited commercial use

Pro Tip: If you like a style but want it in a different genre, feed your prompt into ChatGPT and ask it to rewrite it for rap, orchestral, or lo-fi punk.

PRESENTED BY ASAPP

🌇 Go beyond AI agents with ASAPP

The Rundown: ASAPP solves the toughest problems in customer service with generative AI. The GenerativeAgent® platform was purpose-built to handle complex, multi-turn customer conversations with enterprise-grade performance, safety, and control.

Join ASAPP’s Beyond AI Agents webcast on Sept. 18 to learn:

• What future customer experiences will look like with AI agents

• The power of a multi-agent voice experience

• The tools you need to monitor, deploy, and test AI agents safely

• Lessons gained from real-world enterprise deployments

Save your spot to be ready for the next era of CX.

AI RESEARCH

💊 Harvard’s AI helps reverse disease in cells

Image source: Ideogram / The Rundown

The Rundown: Harvard Medical School researchers just developed PDGrapher, a free AI model that pinpoints gene and drug combinations capable of transforming diseased cells back to healthy states.

The details:

• PDGrapher examines how genes, proteins, and cellular signals work together instead of testing one drug target at a time, like traditional methods.

• The tool outperformed competing AI systems by 35% when tested across 19 cancer types and delivered answers 25 times faster.

• Researchers validated the tool by having it predict known lung cancer treatments, which it correctly identified, along with promising new targets.

• Harvard teams are using the tool to find treatments for brain diseases like Parkinson's and Alzheimer's through partnerships with Mass. General Hospital.

Why it matters: Most drugs today work by hitting a single target in the body, but complex diseases often outsmart the one-trick approaches. PDGrapher's ability to find multiple pressure points simultaneously could crack diseases that have eluded treatments, while potentially saving billions typically lost on dead-end drug trials.

QUICK HITS

🛠️ Trending AI Tools

• ⚙️ Replit Agent 3 - Autonomous agent for building production-ready apps

• 🤖 Qwen3-Next - Alibaba’s hyper-efficient 80B parameter hybrid model

• 👋 Poke - Intelligent AI assistant embedded into text messages

• 🧠 Oboe - Instantly create courses to learn about any topic with AI

📰 Everything else in AI today

Rolling Stone parent company Penske Media Corp. sued Google over the tech giant’s AI Overviews, arguing it illegally uses the publisher’s work while hurting traffic.

Apple senior AI executive Robby Walker has reportedly left the company, having previously worked on both Siri and an AI web search system.

OpenAI’s new agreement with Microsoft will reportedly drop its revenue share from around 20 percent to 8 percent by 2030, saving the company over $50B.

xAI is reportedly laying off 500 generalist AI tutors from its data annotation team, responsible for training Grok AI.

Google’s Gemini overtook ChatGPT as the top iOS app in the U.S., after its Nano Banana image model went viral with people creating their 3D figurines and more.

China’s Tencent just poached Yao Shunyu, a prominent AI researcher at OpenAI, to work on integrating AI into its services.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from an Anonymous reader in Iceland:

"I learn Japanese. I memorize the 2136 regular-use kanji with mnemonics. I use Gemini to analyze the mnemonics, ensuring they are specific, consistent, and systematic. Once I have the mnemonic, the machine turns it into an image prompt. That result goes to a variety of tools, from which I pick the best result. Recently, I experimented with editing these images in Nano-banana, trying to place them in photographs of familiar places, and to make a memory palace."

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: The world’s first AI cabinet member

• Read our last Tech newsletter: Pharma's $1B bottleneck meets AI

• Read our last Robotics newsletter: Zoox' driverless robotaxi service

• Today’s AI tool guide: Create enterprise-ready sounds w/ Stable Audio 2.5

• RSVP to next workshop @ 4PM Friday: Build reliable AI assistants

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer—the humans behind The Rundown