znnajdla

We need something like this and I have been thinking a lot about how to build it. Zulip is closest to what I want but quite untuitive to new employees. It would be fantastic to have one universal app to talk to Claude Code over SSH, employees in Slack-like team chat, and customers and external people over regular email. I want the semantics of email (thread based conversations, inbox zero, universally works with everything) with the UX and speed of Telegram. The protocol I am trying to design is something like SSH that gracefully degrades to POP3/SMTP/IMAP.

Shit that 99% of CRUD app SaaS don't need.

What was the point of the Space Race between the US and the USSR? Proving who had more powerful capabilities. There doesn’t need to be any near-term profit at all. China has a lot to gain by beating US AI even if it makes no money from it. Money is such a small concern compared to the prestige and strategic benefits of being the global center of intelligence research.

To answer your question, what can China do if it wins? Become a global talent/education magnet, replacing the US, for example.

given the enormous costs

The costs are enormous only in America. The actual cost is much lower. American technogy in general is ridiculously overpriced -- compare the cost for raw compute on AWS versus Hetzner for example.

What's the incentive for the Chinese labs to continue releasing weights 5 years from now?

Engineers love to build things (just look at the whole world of open source), and building an AI model is one of the most exciting things to work on. It only takes a few million dollars of funding to produce an AI model. People spend millions on artwork and paintings for fun and prestige. I can easily see why a billionare or government would want to have their own AI model, but is not interested in the business of selling it, so they release it for free. Combined with engineering talent that loves a meaty problem like AI, I see absolutely no reason why open source AI will not continue to thrive, and not just in China.

AI models cost tens of millions to train.

There are people that spend tens of millions of dollars on paintings and artwork. I can see plenty of reasons why organizations and individuals will continue to want to drop a few million on an AI model just for the fun and prestige.

I'm sort of baffled by what the entities that train the open-weights models get out of it though.

Why is it so baffling that people want to build great things? There are plenty of people who are happy building things for a salary and have no interest in taking over the world. Do you find the whole world of open source software baffling? Linus Torvalds and Richard Hipp and Antirez created the world’s most prolific software products and released it for free.

I have a hunch that OpenAIs proprietary compaction endpoint actually does something similar on the backend.

Please update the blog post to clarify. “Claude” is not a model, and your writing makes no sense without specifying.

If you think your codebase is your competitive advantage or trade secret, you're in for bad time. AI coding has made code cheap it's the least valuable part of the value add for most companies. Almost any software company can easily be replicated with a few AI prompts. The real proprietary value is in the marketing distribution partnerships and data, not code.

My experience mirrors this: services like OpenRouter that promise “failover” are pretty much useless except for sandbox testing because models in production are not really interchangeable. Any production harness doing serious agentic work in production is dependent on more model-specific quirks than you would expect. And even if another model works without errors, performance and efficiency is a whole different story. Even the system prompt can and should be tuned to a model’s preferred speaking style, for example <xml tags> for Claude-like models because they were trained on it, while other models do better with other delimiters. Think of the whole harness, prompt, and model as one system, not really with modular parts that can be swapped out if you care about optimal performance.

Not sure why you would think a focus on efficiency means less performance. Compression is intelligence. Higher efficiency enables higher performance.

It’s not for the AI inference, it’s for the tool calls and desktop GUI app workloads and browser. There aren’t any on-device models capable enough of real work that can run on lower end Mac Minis. But for running a few browsers and GUI apps, you’re much better off buying a Mac Mini than paying for a more expensive and worse-performing container in the cloud. Browsers were not designed to run in Linux containers but they run optimally on baremetal desktop OSes. An M4 Mac Mini beats the single core performance of any VM you might rent in the cloud, in terms of raw compute per dollar (Geekbench scores).

This was really thought provoking — it made me realize that Git just happens to use a filesystem for persistence, but doesn’t necessarily have to. A POSIX filesystem might not even be the best way to store a git repo. Makes me wonder: what else could speak Git + POSIX? Redis? Postgres? IPFS is a fun one — it’s already content addressed.

I wouldn’t call it gymnastics. The surprising part of the article was that Git itself is an object store that happens to use a filesystem for persistence, but an S3 bucket might actually be more suitable than a .git directory on POSIX.

Very timely as I just set up Forgejo yesterday. There were immediate unexpected benefits moving off GitHub. It’s hosted on an Orbstack VM on a Mac Mini M4 on my local Tailscale network and the UI is incredibly snappy and quick to browse. Because it’s located inside my tailnet with no public access I don’t need to setup authentication, so agents running inside our network can clone a repo with no fuss (I just protect master branch pushes with a config rule). Github auth is a pain to securely distribute, when you self-host the problem just disappears. Also because it’s running inside our tailnet we can use Forgejo actions for infrastructure-related tasks: Ansible scripts that provision or maintain Mac Minis or Linux servers on the same tailnet over Tailscale SSH. Much simpler than Kubernetes which would be overkill for a small homelab/startup. And the Orbstack VM is literally two clicks to backup and restore or move to another host — I setup the VM running Forgejo on my local MacBook then a few minutes later exported the VM image and transferred it to a Mac Mini server in a few clicks. Backup and restore an entire server to a single file — that is the fastest and easiest server migration I ever performed. I literally can’t think of a single reason I would want to go back to GitHub. Forgejo is a joy to use.

For me personally: I sync OpenCode authentication tokens across all containers using Syncthing (will probably get a NAS at some point) and keep it signed it to Codex, DeepSeek, Kimi as well as a local Gemma 4 12B running on a MacMini. For frontier inference API endpoints you can also use vibeproxy to give you OpenAI compatible endpoints to Codex/Antigravity.

Why are databases so hard?

DeepSeek V4 Flash is a very capable coding model that runs well on the hardware you described. Look up the optimized version specifically designed for local use.

Already been done. Look at the Forge project for local LLMs. It can bring 8b models up to Opus-like performance at tool calling.

This is exactly why I chose to build my AI startup outside of the United States. I knew that if I ever built anything of world-changing consequence, the US would seize it for its own use, because that is what desperate, flailing empires do. Having access to the most capital/talent/resources is irrelevant if I can't keep what I build.

It’s neither a matter of profits nor costs. It’s a matter of incentives. The basic principle of capitalism can’t work if the buyer doesn’t have a choice, if the buyer is not the one who usually pays, and if the payer has a way of avoiding responsibility to pay. The privatized health insurance system is truly and completely fucked up.

That’s not clear at all from the docs. It shouldn’t be called “setup” then. Even after doing sem setup there should be a CLI flag to get the default diff output without unsetting up. Very annoying hijack.

Looks very useful, but incredibly obnoxious that it overrides default git diff. And the only way to get the regular git diff output is to uninstall it? There is no way I would ever want to do that.

Not just domain expertise. The hard part has always been marketing, distribution, risk appetite, motivation, grit, and patience.

There are plenty of things which are "trivial" to produce with no moat and yet are still million dollar businesses. Kebab stands. Water bottles. Barber shops. Movers.

The Abrahamic religions have a natural safeguard against excessive wealth concentration: ban interest-based loans to private individuals (usury is shunned in the Bible and the Koran) and instead encourage a wealth tax on hoarding that directly transfers charity from the wealthiest to the poorest (tithe or zakaat).

Some modern economists have suggested this should work theoretically if properly implemented. See Helmut Creutz, Das Geld-Syndrom (1993) and “The Natural Rate of Interest Is Zero” — Mathew Forstater & Warren Mosler.

Instead of empty theorizing, we should have benchmarks for this. There is at least one benchmark which suggests that LLMs are better at writing Elixir than most other languages: see the AutoCodeBenchmark.

This kind of attitude is exactly why American-style capitalism is not wanted. As if money is the only thing that matters? Maybe there aren’t enough Dutch investors with deep enough pockets to match the American offer because the US prints the worlds reserve currency with no fiscal balance whatsoever. American investors are playing with monopoly money propped up by an army to bully anyone who doesn’t take it.

Because very powerful private VCs and investment bankers want to ensure that governments stay impotent when compared to their capital. Welcome to the Western world.

Concerning for you, perhaps. I take the opposite lesson: this incentivizes and invites the exact kind of entrepreneurs you want to have in a country. You profit-driven Americans assume that all entrepreneurs build things to maximize profit at the expense of everything else. No, not everyone is driven by that, and many of the best company builders were not primarily profit driven (e.g. Steve Jobs believed in beauty and excellence). There are plenty of entrepreneurs who want to build things to support the sovereignty and livelihood of the places that they live in while making a profit that doesn’t jeopardize that.

I say, blocking foreign takeover of vital companies would actually incentivize me as an entrepreneur to choose the EU, and I say that as a startup founder in Europe. Because it levels the playing field for founders who believe in sovereignty: now I don’t have to worry about competitors selling out to foreign capital doing better than me due to that.