joss82

In a good way or a bad way?

Hi, Parseur founder here :D

I understand what they are trying to do, but to me it feels like the moment when MongoDB entered the database space, with semi-structured, "flexible" storage format. It has its uses, for prototyping mostly.

But in high-volume, production workloads, giving a structure to the data you extract (what Parseur does through defining the Fields in your Mailbox, basically giving your output data a schema) adds a ton of value, and the larger the dataset, the truer it is.

Usually, you start by defining where you want your data to go, and which structure it should have, before working backwards from here and starting to extract the data. This is the key to automating your document workflow.

I've been working on Parseur for the last 10 years, and OCR has not been solved yet, let me tell you.

OCR still sucks in 2026. Hopefully this might improve the situation but I haven't tested it yet.

There is no mention of the amount of fuel used to transport the fuel in the article. From what I know it’s a tiny fraction: boats are efficient at transporting stuff (slowly)

That is not what I understood from the article. What I understand is:

Fossil fuels are 40% of freight tonnage, but transporting them fuels is responsible for 50% of the total freight fuel consumption.

I assume 99% of freight uses fossil sources as fuel.

Yes, this is also my feeling.

It happens for every single Anthropic release. Then I try it on real dev and the result is laughably bad. Except in design where it has been doing a decent job for a while. I am not a designer and my bar is pretty low.

Thank you for this answer, it saved me hours of experimenting.

And bonus points for bashing MongoDB, of course. Every single project that I worked on where MongoDB was used, also had MongoDB as the single largest constraint and operations time sink.

Dude. You are marketable as fuck.

Don’t live in the hype. Not everyone is drinking the ai cool-aid bottoms up.

What do you mean by fastapi being a mistake ?

Are you building the site from the same json files that are used in the game?

AIs are still computer programs and are not given the resources to render javascript, so they cannot access the game data from the website. And they obviously don't have it in their parameters. BTW Google Gemini Pro just told me that they know the game but did not know the value. Actually, it points out that it is a known trap for AI to confidently give a wrong, hallucinated value. Maybe it had already seen this thread and integrated it in their parameters or fine tuning. I don't know...

Gemini Fast confidently gives me a wrong value. But very quickly! I'll attach the entire Gemini response as a sub-reply.

They swabbed them and sequenced their DNA.

Parseur | Front-end dev with design appetite | REMOTE [GMT:GMT+6]

Parseur is a B2B Saas that automates document-processing workflows.

We are a 100% office-less, fully remote team of 5 looking for the 6th full-time (or 80%) team member.

We're looking for an excellent front-end dev who ideally knows React. And it would be nice if you could also design and know your way around CSS.

More info here: https://parseur.com/jobs

You are not alone.

After falling in love and hacking away with Claude for a few weeks, I'm now in the hangover phase, and barely using any AI at all.

AI works well to build boilerplate code and solve easy problems, while confidently driving full-speed into the wall as soon as complexity increases.

I also noticed that it makes me subtly lazier and dumber. I started thinking like a manager, at a higher-level, believing I could ignore the details. It turns out I cannot, and details came back to bite me quickly.

So, no AI for me right now, but I'm keeping an eye out for the next gens.

Terragon is amazing. I actually ordered a second laptop to be able to work on several Claude Code PRs at the same time, but with this I don't even need the extra laptop! I've been trying it for 24 hours and it's a nice boost to my productivity, on top of Claude Code's own boost. Yeah!

You missed the joke ;)

Solar and wind cannot replace base load power. Especially not in Germany. They have to rely on peaker plants even more, and those are burning gas, emitting CO2. And they built more coal power plant units, like Datteln 4: https://en.m.wikipedia.org/wiki/Datteln_Power_Station

If it’s cooled in an “open cycle”, it means that the water vapor is released in the atmosphere, via these huge aero refrigerators towers. It will eventually fall back down as rain or snow. Water is not a scarce resource in Germany. Shutting down those plans was an ecological and economical disaster bordering on high treason.

Interesting concept. Let's run the numbers...

The largest Q-Max-class gas tanker is 345 meters long [1]. Let's say you manage to fit 3 giant Siemens wind turbines on it, with 100m long blades [2]. It's a bit cramped but let's say you have extenders on the side to make room for all 3 of them. And also let's say you found a way to prevent the ship from tipping over when the wind is strong. By deploying floaters on the side or whatever. Not unsurmountable.

Each of those wind turbine has a rated power of 14.7 MW [2]. Let's say that you found a place where the wind blows super strong (but not too strong) and steady all the time. It's possible, since you are a mobile ship, after all. Let's say that you have a way for the ship to keep in the same place despite the strong and steady wind pushing you constantly. Using engines is going to lower your efficiency, so let's say we found another way.

So, now your ship is generating 45MW constantly. According to ChatGPT, this is 32 kg of hydrogen per second, taking hydrolysis losses into account.

Tanker capacity is 18 620 000 kg of liquid hydrogen. It will take 581 000 seconds to fill up. 9697 minutes, 161 hours, or 6.7 days. Much shorter than I thought... Did I miss something?

[1] https://en.wikipedia.org/wiki/Q-Max [2] https://www.offshorewind.biz/2024/04/22/first-siemens-gamesa...

Texas yard is the size of a small country

Yes, you are right that playing a game with infinite ammo and health is boring.

However, this never happens in the real world: You buy a nice car, then a nice house, then go to nice vacations, eat better, healthier food.

Then you realize that there are layers of nicer everything above what you consider "nice". Build a more ambitious company, and end up building rockets to explore the solar system and make humanity multi-planetary, cure illnesses, extend knowledge or education or peace.

I spent the last few days testing Llama3 on different GPUs, to find the cheapest cost per token. Spoiler: it's the Nvidia L4, surprisingly.

It's a bit like trying to know if a geometric series is going to converge to 0 or diverge towards infinity.

If a blunder is a 0, then avoiding blunders is super important.

For example if you are in finance or accounting, commiting fraud makes you lose your license and set your business value to zero.

This only works where the number of moves is finite, and you can only iterate in lockstep with your opponent: tennis and chess are good examples.

If you are a startup, you don't have to wait for your competitors to play before making a move, you can (and must, to survive) make as many moves as you possibly can, to get ahead.

A blunder is not as bad as not making enough great moves.

Haha

Neural Networks: from zero to hero

https://karpathy.ai/zero-to-hero.html

It must be "The Grug Brained Developer", by far.

https://grugbrain.dev/

What is yours?

`ctypes` is part of Python's standard library and allows you to directly call C functions from Python code.

It's glorious in its simplicity.

https://docs.python.org/3/library/ctypes.html

Python won because it's a good and versatile programming language, in the sense that it can do a lot of things well. AND it's the fastest language to learn, if you know English, because it's the programming language closest to English that we have. Until AI allows us to program in English (or any language for that matter) directly.

I started doing Python 15 years ago for game development (using Pygame) and it was fairly good at it. At least for prototype, side-project games. Coming from C, the syntax wasn't hard to grasp.

Then I tried it for web, coming from PHP. It was good at it too. I could reuse some of what I learnt in the previous step.

Same for AI, now I can dig into it and only learn the AI part. I don't have to learn yet another new language. And it does the job here too.

Python won because it has the least amount of friction at every step.

Simulating playing cards is surprisingly hard.

Doing so flawlessly at consistent 60fps in a mobile browser is quite the feat, don't you think?

Is it the same car, though?

For efficiency, newer cars have become more profiled and glide through the air where older cars were more boxy.

Those are not "too smart" engineers. Those are "too dumb" engineers, with a will to prove that they are not that dumb to the rest of the company, or themselves.

Explain them that complexity is their worst enemy.

They won't understand.

Explain them again, with more examples, real life situations where complexity wasted time, money, and even, lives.

If they keep not understanding, fire them.