Show HN: AI generated puzzles from Wikipedia articles

https://doodleai.darraghoriordan.com/

by darraghor • 4 years ago

94 13 4 years ago

doodleai.darraghoriordan.com

I suppose the answers could be fed back into the model, to better learn what prompts actual people expect corresponds to a generated image.

doodle:ai - Game #24 - 3/5

obert• 4 years ago

isn’t the top score 4, as in N out of 4?

doodle:ai - Game #22 - 4/8

doodle:ai - Game #17 - 4/7

I don’t get it, 4/8? I see there are 8 valid answers but you can select only 4, or is there a time limit?

darraghor4• 4 years ago

Hey! Thanks for playing the game.

This ratio is describing the game shape (the games are automatically generated and they are all different depending on the wiki article content). 4/8 means that you needed 4 answers out of 8 for example. Having to select 8/8 is too hard and not fun for a game imho.

Some games have only 2 "answers" in total, so then i make you select all the correct answers, otherwise it's too easy. So the game shape would there would be 2/2.

There may be bugs of course. I'll take a look at that number later. I agree it's a bit confusing but I was trying to make the game more usable. It's tricky when the content is dynamic.

dar

darraghor4• 4 years ago

Yes! i think games like this could be useful for training aesthetic and relevancy output from the models.

Actually implementing that kind of feedback is beyond my skills with machine learning though!

8n4vidtmkvmk• 4 years ago

save the data and give it to someone else who can use it? just keep it anonymous.

baxtr• 4 years ago

It's getting absurd. Now we are training humans on how to instruct AI better. Are we becoming ourselves AI now?

PS: Great game! I recommend you get a separate domain for it. Makes marketing so much easier.

Konohamaru• 4 years ago

This is an extremely fun game. Thank you!

darraghor4• 4 years ago

thanks for playing! glad you like it

don-code• 4 years ago

I'm deeply reminded of MindMaze, which shipped with old versions of Microsoft Encarta in the 90s. While I'm sure the questions were canned and not programmatically generated, it behooved you to go read up on the subjects that it was being asked about, using none other than Encarta itself.

ComputerGuru• 4 years ago

I had completely forgotten about this until you mentioned it! Perhaps the first screenshot (excluding the cover art) at this link will jog someone else’s memory: https://www.kotaku.com.au/2020/07/encarta-mindmaze-94-95-the...

I have such a soft spot in my heart for Encarta. It really was “knowledge at your fingertips” and so approachable for people of all ages. I archived a copy of the last release (2004, I think?) to share with my own kids someday.

Encarta is so engrained in my mind that when Azure announced a new comprehensive security portal or whatever called “Microsoft Entra,” each and every time I would log in to the Azure Portal and see the banner for that, I would do a double-take and think it was some new knowledgebase-as-a-service offering Microsoft came up with and dared to recycle the venerable Encarta brand for.

directionless• 4 years ago

In a different vein, https://www.redactle.com is a fun wikipedia game

lee101• 4 years ago

this is awesome! i made a conversational 20 questions with ai game https://www.addictingwordgames.com/play-game/20-questions-wi... that mostly shows off how easy it is to build AI/text generators with https://text-generator.io

the supposed game there is to get it to admit you are right about what its thinking, but theres lots of meta games you can play with it and break out from it.

i was thinking some bartering games with characters/making money running a shop type game would be awesome, so much endless useful potential for games/apps to be able to talk to a real life like character.

darraghor4• 4 years ago

I wrote about how it's put together here https://www.darraghoriordan.com/2022/11/27/doodleai-ai-stabl...

Show HN: AI generated puzzles from Wikipedia articles

Related Stories