Want to create your next game levels through natural language 🗣️🎮? Wait no more, we present:

"MarioGPT: Open-Ended Text2Level Generation through Large Language Models".
PDF: arxiv.org/abs/2302.05981

MarioGPT also predicts the player's path! MarioGPT is a finetuned GPT2 model that is trained on a subset of Super Mario Bros levels. To incorporate prompt information, we utilize a frozen text encoder in the form of a pretrained bidirectional LLM (BART), and output the average hidden states of the model’s forward pass.

Looking forward to @iclr_conf! We'll be presenting a few works on combining deep learning and collective systems. 1⃣Poster session Monday (iclr.cc/virtual/2022/p…) on Variational Neural Cellular Automata

https://mobile.twitter.com/risi1979/status/1488994529923538949

@rasmusbergpalm w/ @miguelgondu @SudhakaranShyam Then three papers Friday at the @cells2societies workshop. 2⃣In Goal-Guided Neural Cellular Automata we leverage goal encodings to control cell behavior dynamically at every step of cellular growth. Led by @SudhakaranShyam w/ @enasmel

An Anarchy of Methods

In her excellent AI book, @MelMitchell1 mentions our 2013 AI survey paper (bit.ly/3jYeyzL), which came out of an AAAI symposium co-organized by @jeffclune, @joelbot3000, and myself on “How Should Intelligence Be Abstracted in AI Research”. 🧵 It summarised the discussions by the symposium's attendees and insights from keynotes by @GaryMarcus, Risto Miikkulainen, @AndrewYNg, @pyoudeyer, Georg Striedter, and Randall O’Reilly.

Finally done with my first blog post "The Future of Artificial Intelligence is Self-Organizing and Self-Assembling"!

sebastianrisi.com/self_assemblin…

Covering work from our group and others on the combination of ideas from deep learning and self-organizing systems. Based on work by @enasmel, @JoachimWinther, @zzznah, @Smearle_RH, @togelius, @LouisKirschAI, @SchmidhuberAI, @SudhakaranShyam, @hardmaru, @pathak2206, @yujin_tang, @stenichele, @BertChakovsky, and many others. Looking forward to your comments!

Excited to share our work on Morphogenesis in Minecraft! We show that neural cellular automata can learn to grow not only complex 3D artifacts with over 3,000 blocks but also functional Minecraft machines that can regenerate when cut in half 🐛🔪=🐛🐛

PDF:arxiv.org/abs/2103.08737 Very proud of the team @SudhakaranShyam , @DjordjeGrbic2, @Sylvia_Sparkle, @AdamKat0na, @enasmel, @claire__aoi. Hopefully of interest to @kenneth0stanley, @Smearle_RH, @pyoudeyer, @BertChakovsky. Maybe Minecraft+NCAs will be the building block of @gregeganSF's permutation city?😄

@enasmel and myself are excited to announce our paper "Meta-Learning through Hebbian Plasticity in Random Networks" arxiv.org/abs/2007.02686

Instead of optimizing the neural network's weights directly, we only search for synapse-specific Hebbian learning rules. Thread 👇 Starting from completely random weights, the discovered Hebbian rules enable an agent to navigate a dynamical 2D-pixel environment; likewise they allow a simulated 3D quadrupedal robot to learn how to walk in around 40 timesteps in the absence of any explicit reward.