smearle (@smearle_rh) 's Twitter Profile
smearle

@smearle_rh

PhD candidate @nyugamelab. @Lojijuice w/ @MarkoCindric. github.com/smearle safesounds.ca/loji-forest, @degrassi S08-12

ID: 22309973

calendar_today28-02-2009 23:49:18

8,8K Tweet

30,30K Followers

1,1K Following

Julian Togelius (@togelius) 's Twitter Profile Photo

Have you played Connections? It's a really fun word game. Given that I'm posting this link, you would guess that we did something new with Connections and AI and are running some kind of study. You would guess right. We will mine you. Still, a good excuse to play some Connections

smearle (@smearle_rh) 's Twitter Profile Photo

Semantically ambiguous chaos in colorful grid form ftw. My fictional 5 year-old Degrassi baby generalized to held-out rulesets in like 5 minutes.

smearle (@smearle_rh) 's Twitter Profile Photo

Kudos for spotting the reference to Greg Egan's Permutation City. We tried to pick a piece of fictional tech that was not *inherently* evil but uhhh... all lofty creations have unforeseen consequences I guess? en.wikipedia.org/wiki/Permutati…

Kudos for spotting the reference to <a href="/gregeganSF/">Greg Egan</a>'s Permutation City. We tried to pick a piece of fictional tech that was not *inherently* evil but uhhh... all lofty creations have unforeseen consequences I guess? en.wikipedia.org/wiki/Permutati…
smearle (@smearle_rh) 's Twitter Profile Photo

Had the pleasure of reading a draft of this, a great and refreshing overview on the subject. (Also helped compile the index, taking special care to note every animal reference, resulting in a veritable Noah’s Arc of variably cognizant lifeforms.)

Julian Togelius (@togelius) 's Twitter Profile Photo

I've been working on ways of algorithmically generating complete games for most of my career. This new project, led by my PhD student Graham Todd, has some of the most exciting results yet, showing that we can generate not only playable but interesting and enjoyable new board

smearle (@smearle_rh) 's Twitter Profile Photo

In 𝘔𝘢𝘬𝘪𝘯𝘨 𝘕𝘦𝘸 𝘊𝘰𝘯𝘯𝘦𝘤𝘵𝘪𝘰𝘯𝘴, led by The Timerino, we develop a method for generating New York Times Connections puzzles with large language models. By iteratively prompting LLMs to generate one group at a time, and prompting it to incorporate some deceptive

Julian Togelius (@togelius) 's Twitter Profile Photo

My lab had been working on scaling up environment/level generation via reinforcement learning for a while now, and we have some nice results to share with you! PCGRL+, our updated PCGRL framework, is not only an order of magnitude faster but can also scale to larger levels than