Show HN: A simple visual game that LLMs can't solve well

Re:Show HN: A simple visual game that LLMs can't solve well by gus_massa on Thursday October 08, 2026 @02:14PM (#50009681)

Interesting. I expect LLM to soleve it soon, but let's take that win for now.

Does each person get the same photos?

[spoiler alert]

1: a cat. Fine.

2: a black dog. I didn't understand the art of the dog and place it very wrong.

3: A photo with multiple white dogs. I thought it was only one dog and matched the wrong one. I counting it a win for me.

4: Another cat. Almost exact match.

5: A butterfly. I got confused and and place it very wrong too.

[ Parent | on HN ]

Re:Show HN: A simple visual game that LLMs can't solve well by abbassix on Thursday October 08, 2026 @02:22PM (#50009796)

There are thousands of images and thousands of different crops from the images. Some are not solvable, but on average you can solve most of them!

[ Parent | on HN ]

Re:Show HN: A simple visual game that LLMs can't solve well by gus_massa on Thursday October 08, 2026 @05:15PM (#50012407)

For the world record, it would be nice to have the same daily challenge for everyone, but I actually like to lay a few games the same day.

I played like 3 games and I already saw a few repeated images (in different games).

There was a very old post in HN about "Cats as placeholder images" and one trick they used was to crop it to the correct size using the eye as the center, to ensure the part of the image was "interesting". Eyes make images interesting.

I only can beat Grok :(

[ Parent | on HN ]

Re:Show HN: A simple visual game that LLMs can't solve well by abbassix on Friday October 09, 2026 @02:16AM (#50016653)

The repeated images might be due to a ranking algorithm that samples random images. Since not all crops are solvable, images where many people achieved a decent score end up ranking higher.

[ Parent | on HN ]

Re:Show HN: A simple visual game that LLMs can't solve well by gus_massa on Friday October 09, 2026 @12:09PM (#50022602)

I think it's a good idea to rank the images. Anyway, I expect repeated cases to be easier for people and skew the results.

[ Parent | on HN ]

Re:Show HN: A simple visual game that LLMs can't solve well by abbassix on Friday October 09, 2026 @03:39PM (#50025655)

Unless, of course, only the first attempts count toward the leaderboard.

[ Parent | on HN ]

Re:Show HN: A simple visual game that LLMs can't solve well by abbassix on Friday October 09, 2026 @03:40PM (#50025663)

By the way, the server is down until Monday, when we'll be able to reach the office.

[ Parent | on HN ]