ojkelly

I’ve been developing a language for a few years, and even with incomplete semantics and a simple one page example LLMs don’t have much trouble writing it.

I think the language/syntax has an impact, but the tooling around it will be most important for LLMs, in the same way it is for humans.

I bake all my own bread, mostly due to food allergies but also because there is nothing better than a fresh loaf of bread just out of the oven.

My experience and conclusion was largely the same as the author. I found the brevity of many recipes frustrating, which led me to study the details of bread making.

The type of four, how much yeast, salt and water. How the water amount varies for so many reasons in a home kitchen.

Making 4 loafs a fortnight gives plenty of opportunity for iteration and learning. And over time I was able to develop a good understanding of dough hydration and how that affects the end product, and how it can make kneading and handling the dough easier or harder. There is a sweet spot.

This also led me to adding oil to a dough. Which acts to keep the gluten chains from getting too long and chewy, thus making the dough short.

It’s also how you make a great pizza dough.

From there I started experimenting with different flours, because I had to if I wanted bread. Running low on bread flour I added semolina, and also made a pure semolina loaf.

The main thing I’ve learned is there is a lot of leeway in the recipes that still produces tasty food. I’ve made some incredible and some woeful loafs that all taste great.

I still prefer specific recipes in metric, but I find more and more I’m reading into what the ratio of ingredients is doing and why.

Until such a point where have agents not trained on human language or programming languages, I think something that’s also really good for people as well. - one obvious way to do things

- memory safe, thread safe, concurrency safe, cancellation safe

- locality of reasoning, no spooky action at a distance

- tests as a first class feature of the language

- a quality standard library / reliable dependency ecosystem

- can compile, type check, lint, run tests, in a single command

- can reliably mock everything, either with effects or something else, such that again we maintain locality of reasoning

The old saying that a complex system that works is made up of simple systems that work applies. A language where you can get the small systems working, tested, and then built upon. Because all of these things with towards minimising the context needed to iterate, and reducing the feedback loop of iteration.

I thought the same, but when Berkeley Mono got ligatures I gave them a go and never turned them off.

I think the truth is that any good monospace font is designed with an awareness of the grid those characters are laid out in. The rhythm and stability of that grid is a feature of monospace fonts. It lets us line up text, draw shapes and so on.

You would think not having the underlying characters visible would be an issue, but ligatures are just symbols like any other. In a short time you learn to read them, like you would any contracted word.

Spatial Audio for music is interesting and when properly mixed for it a song can be great. But I’m not going out of my way to find those songs.

When it absolutely excels is movies and TV, the immersion is spectacular.

I’m holding out for an example of Spatial Audio/Dolby Atmos where the immersion it can provide adds to the experience. Orchestral music is probably the best place to find that.

Would it make more sense to consider the response from the DB, like a response from any other system or user input, and take the parse don’t validate approach?

After all, the DB is another system, and its state can be different to what you expected.

At compile time we have a best guess. Unless there was a way to tell the DB what version of the schema we think it has, it could always be wrong.

The name should represent what the function does, it should indicate its purpose.

The distinction is useful even when it’s structurally identical to another function.

Two identical functions in different contexts or domains often diverge. When they’re not arbitrarily bound by the fact they have the same contents it’s easy to extend one and not the other.

During compilation, they could both end up using the same actual implementation (though I’m not sure if any compiler does this).

Test names are one of those things that are painful because it’s obvious to you as you write it, but there’s an extra hassle to switch gears in your head to describe what the contents of the test is doing.

It is really valuable when they are named well.

I’ve found this is where LLM can be quite useful, they’re pretty good at summarising.

Someday soon I think we’ll see a language server that checks if comments still match what they’re documenting. The same for tests being named accurately.

It’s a consequence of thinking light/dark modes are personal preference when they’re accessibility features.

And when viewed from that lens you can see how off-putting it is when it’s a paid feature, or only adjustable when logged in, or a user-setting that seemingly always defaults to light.

One practice I really like with Feature Flags is having a flag budget. Building on the idea of an error budget, where some percentage of errors is okay, but at a certain threshold feature work needs to slow or stop to get it under control.

Having a limit of in-flight feature flags, means you’re incentivised to clean them up when they’re done (or decide they’re long running features etc), and it can help keep a handle on in progress work.

But mostly, feature flags are a powerful tool—they can hurt as much as they help if you don’t use them right.

There’s 2 mains ways to do feature flags.

Coupled to a code deploy or not.

Changing flags on deploy is simple, great when you have a fast pipeline and only a few flags.

But at some point it becomes useful to decouple feature release from code deploy.

And the only way to do that, is to be able to change the value of a flag out-of-band of a pipeline.

Then you have the capability to test new code in environments before prod and in small parts of prod—canary releases and so on.

Configuration and settings management overlaps with feature flags, but note that often the value comes from the ability to test and safely deploy new code into production environments (more of a release flag), that to enable a feature for a specific user. It just so happens that the use cases and technical implementations overlap so frequently it’s sometimes less work to use the same system.