That’s amazing.

Running Claude Code in your own container with happy is a good way to achieve that. happy.engineering

It’s not the doorman fallacy if removing the driver works. The fact that many new metro systems are driverless, without issues, and that older systems successfully automate, proves it’s not the doorman fallacy.

If you want to know more about why California resisted building housing (spoiler: racism) and why it's finally starting to, you should read this piece by @jetjocko.bsky.social www.motherjones.com/politics/202...

But then again a lot people felt the Biden economy was bad.

It’s not the lump of labor fallacy all over again. Productivity gains translate into doing more with our resources and time not working less.

The instruction set is different, and undeterministic, but it’s still engineering.

Getting an LLM to do what I want within the degree of tolerance I can accept is an engineering problem.

There’s plenty of intention and authorship in driving the system towards your goals from a higher level. Like writing a program in a higher level language is not less intentional or authored than writing assembly.

It’s one of my favorite reads of these past few years. Also gifted it multiple times.

My fave instruction: push the umami to dangerous levels.

Incidentally, I’ve also learned how umami works.

Now when I try something new I go through a few iterations with ChatGPT grounded in that project. Usually takes 4 or 5 rounds of back and forth dialogue to get to a point I’m happy with. I start with an idea, iterate with ChatGPT.

I’ve taken pictures of all my cabinets and got ChatGPT to generate a csv of all my spices and condiments. Then I’ve ran a few deep research on topics like umami, veg takes on French cuisine… and saved them in a project.

Yeah we don’t need formal verification we need to reduce the variance of the output to be within tolerance. Already putting codex as a verifier to claude code is a step change in output quality.