Twitter Feryal Adaptive Reinforcement with XLand games, rather model human groups and speech
Feryal Behbahani at https://arxiv.org/search/cs?searchtype=author&query=Behbahani%2C+F
Edan Meyer @ejmejm1 I totally forgot to post my recent video on AdA: https://youtu.be/BkWLCrLapQo
Check it out if you want to see how @FeryalMP & colleagues successfully scale RL and pave the way towards future RL foundation models. It’s some great work!
Human-Timescale Adaptation in an Open-Ended Task Space at https://arxiv.org/abs/2301.07608
@FeryalMP Watched your AdA video, read the paper. Suggest you not play games and model your group, its interactions, goals and rewards at scale. Model your tools and their capabilities. NOT by you doing it. Teach the AIs your job(s). Let them run the software. Try part of speech.
If they model the jobs of each individual member, then those AIs can do more of the AI research themselves. Automate running software used in AI research, to let humans be more creative and work on deeper problems, devise new industries, create new jobs.