Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Submission - 400 Puzzles Over the Weekend #4

Open
nicogs opened this issue Sep 9, 2024 · 2 comments
Open

Submission - 400 Puzzles Over the Weekend #4

nicogs opened this issue Sep 9, 2024 · 2 comments

Comments

@nicogs
Copy link

nicogs commented Sep 9, 2024

After doing the 6 puzzles on the ARCPRIZE homepage, I had the "How hard can it even be?" and thought of ways certain variables should be passed along the neural network and others could be learned to recognise. For example that 4 blocks of a certain color in a certain position could be a square or a line, isn't a hard job. My plan was to do a few of the hard puzzles and take note of the variables and or meanings of certain things, I started with writing down: difference between Input and output for each test and the differences between the tests themselves as the things that they have in common.

As history repeats it's self when participating in these challenges: I got carried away by the puzzles themselves. Friday at 19:40, I started recording my adventure. My submission this week has not a single line of code nor did I write a single line of code. And there is noting to try (other than trying to solve the puzzles yourself). So what is my submission?

Well, I managed to solve all of the 400 hard puzzles over the weekend, manually. And plan to upload the full (redacted tho) recording when edit&render are finished coming days. Knowing WHAT the solution is for a puzzle, won't get you very far. The HOW is far more important. Seeing someone solve the puzzle and how they tried different wrong ideas will make it easier to teach an AI to solve it, in the end.

Here is a small 15 second timelapse of the whole weekend

I did also learn a few valuable lessons doing this:

1. I should Look at the different tools and get up to speed on what is possible within a certain app.
Only after puzzle 64 I discovered that copy pasting of blocks worked & there was a fill tool. This is why my completion time drastically improved and made les mistakes.

2. As Sherlock Holmes once said: "When you have eliminated all which is impossible, then whatever remains, however improbable, must be the truth."
After double checking if I missed a block, color or rule, completely restart the puzzle several times, getting mad (will be censured in the full version tho). Instead of wasting almost 45min on a single puzzle, I should've anticipated likely mistakes someone could've made and ambiguities examples when creating this puzzle. In the end I found 3 mistakes: (50f325b5, 0d87d2a6)

3. When streaming, avoid music.
I am used to constantly have music on, from a sing along track from the Beatles to darker techno tracks from VTSS, but also Bugge Wesseltoft presenting his New Concept of Jazz or Frank Ocean. Unfortunately you can not use music without the proper licenses when streaming. Apart from the complete ridiculous look of me singing along a track without the track actually playing, I will have to mute most of the audio which also contained explanations for some of the harder puzzles

Some stats:

  • Average completion time: 01m18s
  • Median completion time: 38s
  • Average tries: 1,6
  • Median tries: 1
  • Listened to 19h29m of music
  • Drank 4 bricks of milk, 5L water and 9 cups of coffee
  • Fell asleep 2 times for +-15s behind my desk and yes this is recorded and pretty funny

So I will upload the video with a +18h length in the next few days where I manually find all the paterns, and was able to hit a 100% on a benchmark although the tests contained some mistakes 🙆‍♂️

Cheers

Nicogs

@vjeux
Copy link
Collaborator

vjeux commented Sep 29, 2024

Can you expand on the "mistakes"? They seem to be working fine.

50f325b5
image

0d87d2a6
image

@nicogs
Copy link
Author

nicogs commented Sep 29, 2024

@vjeux the mistakes are also mentioned here & explained them in this Twitter thread

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants