vnglst

Direct API access I’m afraid, it was not my intention to spent it all in one go on this. But after 12€ I didn’t want to stop anymore

Would love a PR with this!

Not sure if it would've gone to the front page of Hackernews with that title! I was also trying to make a little fun about the drama around Mythos/Fable: Even though Fable did this really well, to me it does not appear to be fundamentally different from other top models.

This was my second attempt, I'm still learning! Besides, the wolf was freaking me out.

I also realized this, a quick Google search would’ve told me that this game has been made several times before, also way before I ever had this idea. Apparently it’s a pretty obvious game idea.

Ah well, it’s still fun and it does appear to measure how good AI is in creating these kind of games.

Thanks for that, I messed up copying the links into the article!

Maybe, but using state-of-the-art large language models to solve customer support queries with agentic can quickly use a lot of tokens. What I understood from the talk is that they used agents with limited responsibility and (assumption from me) smaller models, to the make sure the answers were quick, reliable and not too costly.

Yes, that's right, it's https://github.com/vnglst/stacked-game-of-life

The link was broken indeed. Should be fixed now. Glad you like it!

I asked Gemini 3.1 Pro Preview to generate the modern artworks as SVG for my Pelican Art Gallery. I particularly like the rendition of the Sunflowers: https://pelican.koenvangilst.nl/gallery/category/modern

I asked Gemini 3.1 Pro to generate some of the modern artworks in my "Pelican Art Gallery". I particularly like the rendition of the Sunflowers: https://pelican.koenvangilst.nl/gallery/category/modern

Oh that's weird. Are you sure it was the basic link and not the one with count down mode?

The one I shared in the parent is for the New Year's countdown: https://tetris-time.koenvangilst.nl/?mode=countdown&to=2026-...

The basic link should show the current time, even at high speeds: https://tetris-time.koenvangilst.nl/?speed=10

Oh wait, I just tested it on a higher speed (20) and I see what you mean. It's drawing over the existing Tetrominoes! I'm going to max the speed at 10.

The year is coming to an end, and looking back I had a lot of fun building quirky, fun, and insightful visualizations. Agentic coding agents have been hugely empowering for me. In 2024 it was “just” autocomplete on steroids. Now it feels more like conducting, directing the agents to go where I please. Often they’re impressive; sometimes they fail at the simplest tasks.

What worked particularly well here was a test-driven (TDD) approach. I asked Claude Code to focus only on the algorithm for selecting the next Tetromino (the Tetris blocks), starting with tests. After that, things went smoothly, though it still took many follow-up prompts (around 100) to reach the final result.

Thanks I love the suggestions! Great ideas for the next iteration.

As I cannot change the title anymore I think that’s my best option now :-)

Wow this one is great!

Ah thanks for this. The game above is lovely and it’s really similar to what I had in mind (I was also thinking of lemmings!). I see in the other comments below that this idea of mine has been created as a game a lot of times already. Seems like I’m not as original as I thought haha

Made with just HTML, CSS & JavaScript. Inspired by Trainspotting and Gina Trapani's Truisms.

Sorry dead link. New post is up here: https://news.ycombinator.com/item?id=43298945

Shepherd's Dog is a game I've wanted to create for a long time, but I never got the sheep flocking behaviour just right. The goal of the game is to herd all the sheep into the pen before nightfall. I've asked several models to create this game and I'm particularly impressed with what Claude 3.7 could do with a one-shot prompt.

- You can play the Claude game here (note: doesn't work on Safari for some reason): https://html-preview.github.io/?url=https://raw.githubuserco...

- o3-mini's version is here: https://html-preview.github.io/?url=https://raw.githubuserco...

Results of other models and a leaderboard is here: https://github.com/vnglst/when-ai-fails/blob/main/shepards-d...

Some videos: https://hachyderm.io/@vnglst/114125938185826311

Shepherd's Dog is a game I've wanted to create for a long time, but I never got the sheep flocking behaviour just right. The goal of the game is to herd all the sheep into the pen before nightfall. I've asked several models to create this game and I'm particularly impressed with what Claude 3.7 could do with a one-shot prompt.

- You can play the Claude game here (note: doesn't work on Safari for some reason): https://html-preview.github.io/?url=https://raw.githubuserco...

- o3-mini's version is here: https://html-preview.github.io/?url=https://raw.githubuserco...

Results of other models and a leaderboard is here: https://github.com/vnglst/when-ai-fails/blob/main/shepards-d...

Some videos: https://hachyderm.io/@vnglst/114125938185826311

Oh no, thanks for pointing this out! I asked GTP-4o to convert the image to text for me and I only checked some of the cards, assuming the rest would be correct. That was a mistake.

I've now corrected the experiment to accurately take the image into account. This meant that Deepseek was no longer able to find all the sets, but o3-mini still did a good job.

Set is a card game where players have to identify sets of three cards from a layout of 12. Each card features a combination of four attributes: shape, color, number, and shading. A valid set consists of three cards where each attribute is either the same on all three cards or different on each. The goal is to find such sets quickly and accurately.

Though this game is a solved computer problem — easily tackled by algorithms or deep learning — I thought it would be interesting to see if Large Language Models (LLMs) could figure it out.

Thanks Joris!

According the the Wageningen University report there is! It's small road, highways but also rail road tracks. The actual report can be found here: https://research.wur.nl/en/publications/landelijk-grondgebru...

This also contains the definitions of land use categories. I've grouped all of the above (roads, rail, highways) under infrastructure.

Thanks! Yeah, getting the right data is challenging, but other than that it’s doable reusing a lot of the current rendering code.

Oh yeah that’s not ideal, that should be a replace! Thanks for pointing this out.

So, that's also what I thought and still think. But the fact the just by adding some randomness to LLMs increases the usefulness of their responses (making them more "creative") seems to suggest otherwise. Maybe creativity can be reduced to just taking a random different point of view and see where that takes you. Much like the contrarian philosophy advocates.

Good catch, glad that LLMs are usually pretty forgiving when it comes to spelling mistakes

Inspired by Andrej Karpathy's excellent YouTube video on tokenizers, I used Llama3 to analyze all of GPT-4's 100.000 tokens and today I wrote a blog post about it! https://koenvangilst.nl/blog/analyzing-gpt-4-tokens

Great feedback, going to take a look thanks!

Sorry folks, it seems like the Global Forest Watch layer has stopped working. That's the layer in pink that was indicating loss of tree cover. Might be the Hackernews effect.