After Sunday’s disturbing events in Sydney, sending a regular newsletter can feel out of step. It is important to acknowledge this before continuing with this week’s update.

👋 Good morning!

Last week's article on robotics in aged care generated some excellent conversations. Thank you to everyone who reached out with their thoughts and experiences.

On a related (and lighter) note, at the aged care expo I recently attended in China, I witnessed the robot uprising firsthand (proof below).

One of the stages had engineering students demonstrating their work on making robots more agile and dextrous. Impressive stuff. Then things took a turn.

Picture 1: Everything under control. Robot walking obediently in front of its handler. All very reassuring.

Picture 2: The little $#/@ made a break for it!

I'm not saying the machines are taking over, but I watched this thing go rogue in real-time. Fortunately, it's only 70cm tall, so I reckon I could outrun it. For now.

What I cover this week:

CHINA INSIGHTS

In several Chinese provinces, aged care organisations receive subsidies of up to $110,000 AUD to implement AI solutions—covering up to 50% of their technology investments. The logic is straightforward: providers face immediate workforce pressures, technology offers solutions, so remove financial barriers to adoption.

Separately, China has deployed a national complaints system where AI analyses calls in real time, surfacing relevant legislation and connecting citizens to the right department. Average resolution time: 42 seconds. Accuracy: 98.7%.

These aren't pilot programs. They're deployed at scale.

The question isn't whether AI carries risks—of course it does. The question is what types of risks we’re comfortable with taking. Read the full analysis on the trade-offs we're making, what baseline standards we actually need, and why the alternative to deploying technology carries its own costs.

Part III of China Insights: An Innovation-First Approach

View Part I and Part II of China Insights

AI IN PRACTICE

I came across a fascinating account from a solo music composer who spent a year treating ChatGPT Pro as his first employee. He paid the yearly $2,400 USD subscription (roughly $3,700 AUD) and used it to build his entire production website, automate competitor research, and handle everything from proposals to business analysis. His expenses dropped from 33% of revenue to 3-5%.

The article gets very specific about actual workflows, but It's an interesting read and something worth considering for aged care operations staff working alongside AI. A few things stand out though.

First, the real investment is time, not just the subscription cost. He spent the year learning how to work with these tools effectively, and that learning curve isn't captured in the dollar figure.

Second, this kind of transformation requires someone who's genuinely curious about how things work and willing to iterate through failures. He rebuilt his website four times in three weeks. He spent hours listening to AI responses whilst walking his dogs. That's not someone ticking boxes on a productivity tool—that's someone obsessed (on a personal level) with figuring out what's possible.

For aged care operations, the opportunity might be in identifying the intrapreneurs on your team—the people already tinkering with better ways to handle their workload—and giving them room to run small experiments with AI subscriptions and protected time to explore.

A Flinders University scoping review looked at digital health adoption across Australian aged care and found a predictable pattern—the providers who'd benefit most from these technologies are the least equipped to use them. The study, led by Dr Naser Pourazad, analysed 48 Australian research papers and found rural and regional services consistently struggle with patchy broadband, limited digital literacy amongst staff, and lack of organisational leadership. Meanwhile, the tools themselves (such as telehealth, remote monitoring, electronic health records) demonstrably reduce hospital transfers, improve medication safety, and help track health changes before they become crises.

A survey of 10,000 South Koreans found that adults aged 60 and over have the lowest current AI usage—only 29% have tried generative AI—but show the highest willingness to adopt it, with over 84% wanting to increase their use, particularly for financial services. Wealthier older adults (top 20% of asset holders) were more confident than younger groups that AI-based robo-advisers will outperform human advisors, citing faster market analysis and accessibility.

Recent AI research challenges one of the most common prompting techniques: assigning your AI a persona to improve accuracy.

The study tested whether telling an LLM "you are a physics expert" actually makes it better at answering physics questions. Apparently it doesn't. Across six different models, expert personas showed no consistent benefit on graduate-level questions. Domain-mismatched personas sometimes made performance worse. Low-knowledge personas were actively harmful.

What actually works: Tell the AI which audience it's answering for, not what role it should play.

Instead of "you are a marketing expert” try "your audience are older aged care recipients with low digital literacy” or "answer this for care workers who are unfamiliar with this policy”.

Specifying your audience changes how it calibrates its response—the technical depth, assumptions, and terminology.

Last week I shared that OpenAI released ChatGPT Atlas, a browser with ChatGPT built in that can open tabs, fill forms, and complete tasks for you. I've since read an interesting article analysing the security implications, and thought I'd share it here. A major analyst firm has recommended organisations block these AI browsers for now, citing security risks that need addressing first.

The concerns centre on default settings that send browsing data (including content from open tabs) to cloud-based AI services, and the potential for autonomous features to make expensive mistakes. Think employees using the browser to auto-complete mandatory training, or AI agents booking the wrong flights when handling procurement tasks. The technology itself isn't the problem, it's whether users and organisations understand the risks and have proper controls in place.

For personal experimentation it's fine (I've been using it myself), but I agree with the assessment; I wouldn't recommend rolling it out across an organisation without serious safeguards.

Every Tuesday I’m looking into some of the changes shaping ageing and aged care, and sharing ideas you can apply in practice. Whether you’re exploring AI tools, rethinking services, or looking ahead to what’s coming, I hope you found something here worth your time.

If this issue was useful, share it with your team or contacts.

See you next Tuesday,
George