I caught Claude the other day hallucinating code that was not only wrong, but dangerously wrong, leading to tasks being failed and never recover. But it certainly wasn't obvious.

I keep explaining to my peers, friends and family that what actually is happening inside an LLM has nothing to do with conscience or agency and that the term AI is just completely overloaded right now.

There is only one source of truth and that is the source code. To define and change contracts written in an ambiguous language and then hope the right code will magically appear, is completely delusional.

Iteration is the only game in town that is fast and produces results.

The democratization ends at your router. Unless you are willing to lay down your own wires - which for legal reasons you most likely won't be able to do, we will hopelessly be dependent on the ISP. (Radio on free frequencies is possible and there are valiant attempts, they will ultimately remain niche and have severe bandwidth limitations)

For decades ISP have throttled upload speeds: they don't want you to run services over their lines. When DSL was around (I guess it still is) in Germany, there was a mandatory 24h disconnect. ISP control what you can see and how fast you can see it. They should be subject to heavy regulation to ensure a free internet.

The large networks, trans-atlantic, trans-pacific cables, all that stuff is beyond the control of individuals and even countries. If they don't like your HTTP(S) traffic, the rest of the world won't see it.

So what you can own is your local network. Using hardware that is free of back-doors and remote control. There's no guarantee for that. If you are being targeted even the Rasperry Pi you just ordered might be compromised. We should demand from our legislators that hardware like this is free of back-doors.

As to content creation: There are so so many tools available that allow non-technical users to write and publish. There's no crisis here other than picking the best tool for the job.

In short: there's no hope of getting a world-wide, free, uncensored, unlimited IP4/6 network back. We never had it in the first place.

No regular user cares about - oh my data, it is stored centrally, how evil! That is just not a problem most people have. Like at all.

No, that is just your interpretation of what you see as something that can't possibly be just token prediction.

And yet it is. It's the same algorithm noodling over incredible amounts of tokens.

And that's exactly the explanation: People regularly underestimate how much training data is being used for LLMs. They contain everything about writing a compiler, toy examples, full examples, recommended structure yadda yadda yadda.

I love working with Claude and it regularly surprises me but that doesn't mean I think it is intelligent.

This is a for-profit company running this service. It ain't free to operate.

If you don't like that, go elsewhere.

If there is one thing that has been a resounding success on the internet it is this: free services that you pay for with your clicks. Just look at the plethora of free services you get.

In no other economy would that be even remotely possible.

Which looks more like a blind spot to me honestly. This category of databases is just fantastic for industries like fintech.

Two candidates are sticking out. https://xtdb.com/blog/launching-xtdb-v2 (2025) https://blog.datomic.com/2023/04/datomic-is-free.html (2023)

Using a schema/spec/type library at runtime that is even more powerful than ADTs is a better investment - or to be less controversial - is an additional investment on top of types.

Yes, it means the compiler can't help you as much, but who has time waiting for a compiler anyways ;)

I find the "make illegal state unrepresentable via types" idea great for software that needs to fly a plane, but for enterprise software not so much. The cost/benefit is not justifiable.

HTMX serves fully baked HTML that needs to be created on the back-end (or front-end-facing servers)

That is more processing than sending the raw data to the front-end and baking the HTML there.

It is also more bandwidth (unless the JSON is more verbose than the HTML generated).

Lastly, I can generate different HTML fragments on the front-end from the same client-side state with the data only being transported once. How is that working out?

The author probably just means LLMs. And that's really all you need to know about the quality of this article.

If that is just your team, use whatever tech gets you there quick.

However, if you need to provide some guarantees to a second or third party with your API, embrace standards like JSON, even better, use content negotiation.

I wouldn't underestimate it - it's the hammer can break up the information silos we've built up around websites/apps.

Why prompt Gertrude(tm) on Ramazon for a specific thing you need, if you can ask ChatGPT to find you said thing along with UIs to purchase it across all e-commerce platforms that agree to play in this market?

There are pros/cons.

It should be obvious by now that using any free service of scale is being paid for by your interactions which are made more valuable through fingerprinting.

Trying to circumvent that just makes it more expensive for the rest of us.

In Clojure, there's no appreciable compilation time. During my work week I barely, if ever, restart the application I'm working on: I'm working inside of it via a REPL connection.

It's an entirely different ball game and if you just compare language features you are missing out on an incredible interactive coding experience.