jhancock

The high level description for Tokyo's management could apply to Shanghai. Replace Tokyo's "elected mayor/assembly" with "party member administrators". Each Shanghai district has its own management structure.

The vague description "run by the central government as a province rather than a city" is uninformative.

Lived in Shanghai 10 years. The city is well run for something of its magnitude. Mostly competent leadership and cultural alignment.

true. I don't have the energy to learn emacs fu. An actual lisp machine would be dreamy.

I use a modern Lisp everyday...Clojure. My dev environment is VS Code using the most excellent Calva extension which give me REPL-everywhere in my editing experience.

Yes, that 70's experience was better...and it's still here, refined and built on modern tooling and runtimes.

sure. It may be more than that...possibly due to variable operating params on the servers and current load.

On whole, if I compare my AI assistant to a human worker, I get more variance than I would from a human office worker.

I can use GPT one day and the next get a different experience with the same problem space. Same with Gemini.

Love where this is going. Your company page https://www.zabaca.com and https://www.zabaca.com/careers/ feel just right.

Because shareholders want "number go up".

Would be nice to know why Apple didn't throw its war chest at the WarnerBros/HBO acquisition Netflix just did.

I spent a few years leading dev on decentralized exchanges, building bridges to other chains and building a sophisticated margin system on top of the trading pools.

A few things I think I've learned:

In its current state, most retail investors are simply supplying to the sophisticated investor.

Although some DeFi projects make a genuine effort to provide analysis tools to level the playing field, it's not nearly enough.

The safest least volatile yields in DeFi are lending your stable coins into a system such as aave. The yield is not far from a high yield USD savings account.

Exchanges such as Uniswap may be the most important legit tool in DeFi. The biggest problem is the liquidity provider's ability to protect their downside...so the investor adds on more sophisticated monitoring/hedging schemes. This gets us back to the retail investor being at a severe disadvantage.

Thats Clojure...one person can look like a whole team ;)

couldn't help being cheeky...actually there are quite a few contributors.

Is there a terminal AI assistant that doesn't have heaps of depenedancies and preferably no node? Claude and codex both require node. I'm a fan of the lightweight octofriend. But also node. I do not like installing node on systems that otherwise would not require it.

containerize all the things...Nix, Podman, Docker. It's not a big hassle once you get through the initial steps.

Would be good to see projects (like those recently effected) nudging devs to do this via install instructions.

China has in recent years built apartment blocks for the unhoused. From pics I've seen, the units are one room studio like apartments. I'm told they rent for around $15 per month. It seems to work for a large set of the unhoused. China's leadership has a mantra of "no welfarism". This is an example of how they achieve it...by enabling even the poorest to have a way to support themselves.

Good article. Thanks for sharing.

Maybe I'm doing it wrong. I've been using AI tools most days the past 2+ years. Occasionally I get good results. I spend heaps of time rewriting/reviewing.

Vibe coding can work if both the programming techniques and problem domain are well understood by the LLM. For the work I do, this means front end stuff.

Back end stuff is where the problem domain sits. I spend so much time explaining the problem domain to the LLM that its best I just write it all myself instead of cleaning up the piles of code the LLM will spit out. So no vibing on the back end...just crack open the AI assistant for debugging.

I don't see companies replacing $150/hour programmers with AI. Not yet. I think what we are seeing is companies spending heaps of money/attention on AI to the point of not making hiring decisions on programmers.

Just prior to the first dot com boom, companies were reinventing themselves with systems replacing those built in the 60s through 80s. These new systems were sophisticated and if done right, game changers. The dot com boom hit with very simple tech: click -> load next static page. This consumed all the attention. I think thats what we are experiencing now more so than a clear job replacement.

thanks for the good news. It does exclude foreign currency accounts.

As 'numbsafari said below, you should no longer user your host for dev..this includes all those cool AI assistant tools. You need to containerize all the things with runpod or docker

New Zealand banks are horrible. If they made it easy to hold USD and/or forex/fees on international card charges were reasonable, Wise wouldn't be needed in most cases.

NZ banks also have no depositor protection. No equivalent of the US FDIC. Note below from 'jemmyw depositor protection added the past couple months.

Supply chain attacks means I can no longer do dev work without a container. On a Mac this means running linux containers which need a base linux VM. I've stared using Podman. It does the job.

On linux this means the same (except the VM isn't required), which gives a more elastic experience for mem/cpu/gpu.

MacOS 26 crosses the line. My next machine with be linux.

they have a Claude Code specific endpoint...see the excellent docs https://docs.z.ai/devpack/tool/claude

I have been subscribing to both Claude and ChatGPT for over two years. Spent several months on Claude's $100 plan and couple months on ChatGPT's $200 plan but otherwise using their $20/month plans.

I cancelled Claude two weeks ago. Pure GLM-4.6 now and a tad of codex with my ChatGPT Pro subscription. I sometimes use ChatGPT for extended research stuff and non-tech.

I tried synthetic.new prior to GLM-4.6...Starting in August...So I already had a subscription.

When z.ia launched GLM-4.6, I subscribed to their Coding Pro plan. Although I haven't been coding as heavy this month as the prior two months, I used to hit Claude limits almost daily, often twice a day. That was with both the $20 and $100 plans. I have yet to hit a limit with z.ai and the server response is at least as good as Claude.

I mention synthetic.new as it's good to have options and I do appreciate them sponsoring the dev of Octofriend. z.ai is a China company and I think hosts in Singapore. That could be a blocker for some.

I've been using GLM-4.6 since its release this month. It's my new fav. Using it via Claude Code and the more simple Octofriend https://github.com/synthetic-lab/octofriend

Hosting through z.ai and synthetic.new. Both good experiences. z.ai even answers their support emails!! 5-stars ;)

How do you review code when the LLM can produce so much so fast?

What I got from the article is the malicious app could read the SMS or email which may contain a 2FA code.

deep-chat is a solid chat UI

https://github.com/OvidijusParsiunas/deep-chat

Good writeup...it's two answers in one ;)

I've come full circle from the first answer given. I wanted to get into neural nets a couple years ago. I chose Clojure because I knew I would be interfacing with components I didn't understand and wanted a system that helped me explore.

thanks! I've tried Plex, Jellyfin etc on my Mac. I've tried three different Apple TV apps as streaming client (Infuse, etc). They are all glitchy. Another key problem is if I want to bypass the streaming server on my Mac and have Infuse on the Apple TV just read files from the Mac the option is Windows NFS protocol...which gives way too much sharing by providing the Infuse app with a Mac id/password.

thanks for the article.

I'm dreaming of this: mini-nas connected direct to my tv via HDMI or USB. I think I'd want HMDI and let the nas handle streaming/decoding. But if my TV can handle enough formats. maybe USB will do.

anyone have experience with this?

I've been using a combination of media server on my Mac with client on Apple TV and I have no end of glitches.

No doubt. The tedious stuff is just a time saver. The domain discovery is often the difference between making progress and giving up.

I'm not a skeptic, but I keep LLMs on a short leash.

This is a thoughtful article. Thanks `tptacek

My LLM use is: 1 - tedious stuff; web pages interacting with domain back end. 2 - domain discovery.

In a recent adventure, I used Claude 4 to tease out parameters in a large graph schema. This is a combination of tedium and domain discovery (it's not my graph and I'm not a domain expert). In the first day, Claude uncovered attributes and relations no other LLM or Google search uncovered. And it worked!! The next day, I allowed it to continue. After a bit, results didn't pass the sniff test.

I checked into details of Claude's thinking: it decided to start making up schema attributes and inventing fallback queries on error with more made up attributes. It was "conscious" of its decision to do so. By the time I caught this, Claude had polluted quite a bit of code. Sure, plenty of well placed git commits helped in rolling back code...but it's not quite that simple..over the many git commits were sprinkled plenty of learnings I don't want to toss. It took another two days of carefully going through the code to pull out the good stuff and then roll things back. So now I'm at day five of this adventure with cleaned up code and notes on what we learned.

I suspect continual improvements on tooling will help. Until then, it's a short leash.

Joy was a key reason I chose Clojure.