Back in Asia
trying to not tunnel-vision
🐙 Hi friend, how’s it going?
I’m back in Hong Kong after New York! 🌸
I did something I haven’t done in years - I actually took a proper break. Visited a monastery, completely unplugged from news, technology, AI, all of it. No feeds, no updates, no keeping up. Just... nothing. And honestly, it gave me a lot of perspective.
I realise I’ve been tunnel-visioning way too hard. I want to be more intentional about how I build and what I build. Operate on my own timeline rather than feeling like I need to keep up with every release and every model drop. 💜
🌻 Interesting Things (I think)
A range of things that I find interesting!
🖊️ Learning
🖥️ Heading to Shenzhen for Hardware
I came back from my break with an itch to build more custom tools and actually get my hands on hardware. Thinking about heading to Shenzhen soon to pick up some parts: mini computers, bits and pieces to integrate with agent frameworks like Hermes. The hardware era is very much continuing. More updates on this when it happens. 👀
🏛️ Governments Intervening on AI - Claude Fable 5
This one’s been on my mind since I got back. For the first time, we’re seeing governments step in hard, intervening on Claude Fable 5 over national security concerns. First real precedent of this kind of intervention because a model is considered too good and potentially too risky.
And honestly, not to be a downer, but where was this energy when it came to people losing skills, early-career hiring slowing down, and unemployment quietly rising in exposed fields? That didn’t seem to trigger the same urgency.
Ah well. If anything, it’s only strengthened my conviction in open source. When governments start pulling levers on proprietary models, the case for open, decentralised AI gets a lot stronger. Just my take. 🐙
🐙 Not Sure Where to Start With AI?
This is one of the most common things I hear: I want to learn AI but I don’t know where to begin. I get it, the overwhelm is real. There’s so much out there.
So our team at Lonely Octopus put together a free 28-Day AI Sprint Roadmap - pick your goal (leveling up at work, building AI products, or offering AI as a service) and it gives you a clear, day-by-day path forward. No more staring at a list of 50 things, wondering where to start.
If you’ve been putting it off, this is a low-commitment way to just begin. 👉 Get it here for free 💜
🎥 This Week’s Video
🐙 Join the waitlist for Lonely Octopus if you’re interested in learning how to build AI agents!
P.S. All constructive feedback is greatly appreciated!
"When governments start pulling levers on proprietary models, the case for open, decentralised AI gets a lot stronger"
So what happens when China clamps down on the open weight models? That's under discussion, as you've probably heard. Foreign access to DeepSeek and others may be restricted as well as future models. Presumably the next step will be forbidding downloads of open weight models. Given that a lot of US startups are using Chinese open weight models, this could become a real problem.
The US government as well as companies like Anthropic seem to be really interested in "Balkanizing" the AI space under the stupid notion of "containing China" - which is impossible.
I have free access to Kimi K2.6 with Nvidia's NIM Developer program. I'd hate to lose that since Kimi is a life-saver when some weird crap happens on my machine. I spend two hours sometimes in troubleshooting efforts. Kimi knows absolutely everything about even my specific Linux distro, openSUSE Tumbleweed. I don't bother going on forums any more for assistance because that takes hours and days, while Kimi can help me troubleshoot in a couple hours.
By the way, if you haven't checked out MSTY Studio, you should. It's a great GUI front end to local and online models with a ton of features including their own version of OpenClaw called MSTY Claw. Check it out: https://msty.ai/
And of course, still bugging you to do a video or live stream on AI-generated code security issues and AI agent security issues. :-)
Interesting, just learned there’s two kinds of AI stacks. There’s your productivity stack, and then there’s a stack when you put a local LLM together, r.a.g, LoRAs. WebUI, postgres, Ollama, LLM Studio, AnythingLLM, Debian, Ubuntu, snaps, flats (wth!?) Granite, Qwen, Gemma, Nemotron, eeesh! I’m doing IBM AI classes. Gemini says I should teach a prompting class. Was wondering what you thought of Prof Hinton, or Frederich Hayek. Seems decentralized AI is the way to go… You give me hope Tina (you and your team) that I might be productive one day with AI 😊 🚬☕️ going back to work now..