Loy

This is great indeed, I installed it a couple days ago.
Turns out I need a bigger computer

Is niet officieel van Google. Was mooi geweest, is het niet

Een gigantische toekan voor de toekangigant!

In de winter wekken we te weinig op, maar dat is in NL met zonnepanelen niet te fixen. Ik zou liever zien dat bijv hoogovens/Tata met mijn overige zonnestroom staal maakt, maar helaas: netcongestie. En ikzou liever mijn energie die ik 's middags over heb 's avonds en 's nachts gebruiken

Voor dat afschaffen salderen is echt wel wat te zeggen hoor. Zonder andere infrastructuur hebben we in NL 'even' genoeg zonnepanelen. De prijs voor energie is deze dagen erg laag of zelfs 0 afgelopen weekend. De focus moet nu gaan liggen op het beter gebruiken van al die gratis energie.

True. The outcome is nearly the same, that is all we can check. I suspect the process (at least how it appears to the artists) is different. But what is really going on...? Unknowable

For coding, it these things make a bunch of mistakes, but that is not to say I wouldn't make the same mistakes.

It's either catch up or be left behind I'm afraid. I don't like that at all, but that is the reality.

Fair point. But I think these things simply work very different from humans. A diffusion process to make an image has nothing to do with intelligence or art: it transforms noise towards a manifold of images we as humans decided we like or find beautiful etc. Same diff between how planes vs birds fly

I have mixed feelings: these things are not intelligent, but they are still useful, at least to me.

This was all done by OpenCode's Zen "Big Pickle" model, I have no idea where it runs. I'd much rather have this running at home, with the waste heat into eg. my water 😅

Anyone wants to play the game? Let me know!

Will we make ever more software because it got simpler (Jevons paradox)? Will we need less people doing that? Will we need more people to clean up the mess for when the agents fails? All of the above?

There is still experience and engineering involved, there is still complexity to be managed, hardware to poke at work, so I don't fear for *my* job per se, but I am very curious to see the impact, especially when these coding agents get even better, which I expect they will.

It's not all roses though, I wanted to post this earlier with some nice screenshots of a game in progress, but of course the whole thing broke while fixing some bugs. While the generated (and sensible looking) unit-tests for the game logic all passed. Testing the web API is on the TODO-list...

Now spending evenings tweaking the thing prompt by prompt. Often in passing; check what it's done for a few minutes before I continue with other stuff I was doing. It's kind of addictive really.

And having a lot of fun! So much that, last Sunday, I had an SSH session open to my laptop from my phone so I could give the agents some hints while I was waiting for my kid to finish swimming class.

This may not be news to a lot of people, but I'm still amazed how well these coding agents work, my mind is thoroughly blown and I'm still piecing it back together.

GitHub - LoyVanBeek/live_battlefield: Play the game Battlefield in real life with others via your phone Play the game Battlefield in real life with others via your phone - LoyVanBeek/live_battlefield

I have no idea what web framework it uses; for this experiment I try to not look at the code at all, but see for yourself at github.com/LoyVanBeek/l.... The AI coding assistent even added an 'AI' player that by the looks of it just throws bombs in random locations.

Some bugs, sure. Oh, and I had no-one to play against for testing, so: "Add an admin panel webpage for testing". Boom, added as well without me touching a single line of Python, JavaScript or HTML.

Threw this description text into OpenCode. Filled a Telegram API key in a config file it prepared. After an hour or so (didn't wait, I was picking out Lego pieces for the kids because Sunday morning...), I have something that I could actually use and play.

Last week Sunday I spent an half hour writing the rules of the game, what domain objects there are, some technology choices, rough outline of how I wanted things set up and basic architecture.

with multiple teams and visiting locations to earn bombs. We played a manually administrated version of this at our weekend for volunteers, last year.

OK, I've drank the Kool-aid, joined the club, had my mind blown, etc: I started 'vide-coding' or 'agentic engineering' as is the fancier terms apparently.
I had some ideas lingering about for cool side-projects, like a Telegram bot to play a game with in real life, for my scout group: Battlefield,

Is die vent niet gewoon moe van al het gedoe en moet ie niet met pensioen ofzo? Lekker achter de geraniums, uit de politiek

Superhippe schrootjes op die laatste foto 😍

We don't want humans, we want superhuman ability. That's why I don't get eg Tesla depending on vision only with the argument that humans use just that. I want super human driving, that probably requires superhuman sensing.

I'm not so sure actually. For the few times per year I would have to shovel snow (or rake leaves, or trim the hedge, or ..., whatever) I could get me a dedicated, efficient machine to do it. My shed would be overflowing. Could also have one robot do all those different tasks. Slower, but all in one

That's really cool! But another showcase that yes, these semi-humanoids are very general but more efficient tools for this applications exist like a snow blower. Same with dishwashing.

However, those are 2 large machines work once every while. That's not so efficient, are humanoids btter that way?

✅ Op de boterhammen
✅ In je slaapzak
✅ Meteen als je wakker wordt aan je voeten

Natural stupidity empowered by artificial intelligence, a disaster waiting to happen

Any other recommendations in that genre 😅?